20220311.md

ArXiv cs.CV --Fri, 11 Mar 2022

1.Conditional Prompt Learning for Vision-Language Models ⬇️

With the rise of powerful pre-trained vision-language models like CLIP, it becomes essential to investigate ways to adapt these models to downstream datasets. A recently proposed method named Context Optimization (CoOp) introduces the concept of prompt learning -- a recent trend in NLP -- to the vision domain for adapting pre-trained vision-language models. Specifically, CoOp turns context words in a prompt into a set of learnable vectors and, with only a few labeled images for learning, can achieve huge improvements over intensively-tuned manual prompts. In our study we identify a critical problem of CoOp: the learned context is not generalizable to wider unseen classes within the same dataset, suggesting that CoOp overfits base classes observed during training. To address the problem, we propose Conditional Context Optimization (CoCoOp), which extends CoOp by further learning a lightweight neural network to generate for each image an input-conditional token (vector). Compared to CoOp's static prompts, our dynamic prompts adapt to each instance and are thus less sensitive to class shift. Extensive experiments show that CoCoOp generalizes much better than CoOp to unseen classes, even showing promising transferability beyond a single dataset; and yields stronger domain generalization performance as well. Code is available at this https URL.

2.Transfer of Representations to Video Label Propagation: Implementation Factors Matter ⬇️

This work studies feature representations for dense label propagation in video, with a focus on recently proposed methods that learn video correspondence using self-supervised signals such as colorization or temporal cycle consistency. In the literature, these methods have been evaluated with an array of inconsistent settings, making it difficult to discern trends or compare performance fairly. Starting with a unified formulation of the label propagation algorithm that encompasses most existing variations, we systematically study the impact of important implementation factors in feature extraction and label propagation. Along the way, we report the accuracies of properly tuned supervised and unsupervised still image baselines, which are higher than those found in previous works. We also demonstrate that augmenting video-based correspondence cues with still-image-based ones can further improve performance. We then attempt a fair comparison of recent video-based methods on the DAVIS benchmark, showing convergence of best methods to performance levels near our strong ImageNet baseline, despite the usage of a variety of specialized video-based losses and training particulars. Additional comparisons on JHMDB and VIP datasets confirm the similar performance of current methods. We hope that this study will help to improve evaluation practices and better inform future research directions in temporal correspondence.

3.An Empirical Investigation of 3D Anomaly Detection and Segmentation ⬇️

Anomaly detection and segmentation in images has made tremendous progress in recent years while 3D information has often been ignored. The objective of this paper is to further understand the benefit and role of 3D as opposed to color in image anomaly detection. Our study begins by presenting a surprising finding: standard color-only anomaly segmentation methods, when applied to 3D datasets, significantly outperform all current methods. On the other hand, we observe that color-only methods are insufficient for images containing geometric anomalies where shape cannot be unambiguously inferred from 2D. This suggests that better 3D methods are needed. We investigate different representations for 3D anomaly detection and discover that handcrafted orientation-invariant representations are unreasonably effective on this task. We uncover a simple 3D-only method that outperforms all recent approaches while not using deep learning, external pretraining datasets, or color information. As the 3D-only method cannot detect color and texture anomalies, we combine it with 2D color features, granting us the best current results by a large margin (Pixel-wise ROCAUC: 99.2%, PRO: 95.9% on MVTec 3D-AD). We conclude by discussing future challenges for 3D anomaly detection and segmentation.

4.AGCN: Augmented Graph Convolutional Network for Lifelong Multi-label Image Recognition ⬇️

The Lifelong Multi-Label (LML) image recognition builds an online class-incremental classifier in a sequential multi-label image recognition data stream. The key challenges of LML image recognition are the construction of label relationships on Partial Labels of training data and the Catastrophic Forgetting on old classes, resulting in poor generalization. To solve the problems, the study proposes an Augmented Graph Convolutional Network (AGCN) model that can construct the label relationships across the sequential recognition tasks and sustain the catastrophic forgetting. First, we build an Augmented Correlation Matrix (ACM) across all seen classes, where the intra-task relationships derive from the hard label statistics while the inter-task relationships leverage both hard and soft labels from data and a constructed expert network. Then, based on the ACM, the proposed AGCN captures label dependencies with dynamic augmented structure and yields effective class representations. Last, to suppress the forgetting of label dependencies across old tasks, we propose a relationship-preserving loss as a constraint to the construction of label relationships. The proposed method is evaluated using two multi-label image benchmarks and the experimental results show that the proposed method is effective for LML image recognition and can build convincing correlation across tasks even if the labels of previous tasks are missing. Our code is available at this https URL.

5.Towards Less Constrained Macro-Neural Architecture Search ⬇️

Networks found with Neural Architecture Search (NAS) achieve state-of-the-art performance in a variety of tasks, out-performing human-designed networks. However, most NAS methods heavily rely on human-defined assumptions that constrain the search: architecture's outer-skeletons, number of layers, parameter heuristics and search spaces. Additionally, common search spaces consist of repeatable modules (cells) instead of fully exploring the architecture's search space by designing entire architectures (macro-search). Imposing such constraints requires deep human expertise and restricts the search to pre-defined settings. In this paper, we propose LCMNAS, a method that pushes NAS to less constrained search spaces by performing macro-search without relying on pre-defined heuristics or bounded search spaces. LCMNAS introduces three components for the NAS pipeline: i) a method that leverages information about well-known architectures to autonomously generate complex search spaces based on Weighted Directed Graphs with hidden properties, ii) a evolutionary search strategy that generates complete architectures from scratch, and iii) a mixed-performance estimation approach that combines information about architectures at initialization stage and lower fidelity estimates to infer their trainability and capacity to model complex functions. We present experiments showing that LCMNAS generates state-of-the-art architectures from scratch with minimal GPU computation. We study the importance of different NAS components on a macro-search setting. Code for reproducibility is public at \url{this https URL}.

6.Prediction-Guided Distillation for Dense Object Detection ⬇️

Real-world object detection models should be cheap and accurate. Knowledge distillation (KD) can boost the accuracy of a small, cheap detection model by leveraging useful information from a larger teacher model. However, a key challenge is identifying the most informative features produced by the teacher for distillation. In this work, we show that only a very small fraction of features within a ground-truth bounding box are responsible for a teacher's high detection performance. Based on this, we propose Prediction-Guided Distillation (PGD), which focuses distillation on these key predictive regions of the teacher and yields considerable gains in performance over many existing KD baselines. In addition, we propose an adaptive weighting scheme over the key regions to smooth out their influence and achieve even better performance. Our proposed approach outperforms current state-of-the-art KD baselines on a variety of advanced one-stage detection architectures. Specifically, on the COCO dataset, our method achieves between +3.1% and +4.6% AP improvement using ResNet-101 and ResNet-50 as the teacher and student backbones, respectively. On the CrowdHuman dataset, we achieve +3.2% and +2.0% improvements in MR and AP, also using these backbones. Our code is available at this https URL.

7.LoopITR: Combining Dual and Cross Encoder Architectures for Image-Text Retrieval ⬇️

Dual encoders and cross encoders have been widely used for image-text retrieval. Between the two, the dual encoder encodes the image and text independently followed by a dot product, while the cross encoder jointly feeds image and text as the input and performs dense multi-modal fusion. These two architectures are typically modeled separately without interaction. In this work, we propose LoopITR, which combines them in the same network for joint learning. Specifically, we let the dual encoder provide hard negatives to the cross encoder, and use the more discriminative cross encoder to distill its predictions back to the dual encoder. Both steps are efficiently performed together in the same model. Our work centers on empirical analyses of this combined architecture, putting the main focus on the design of the distillation objective. Our experimental results highlight the benefits of training the two encoders in the same network, and demonstrate that distillation can be quite effective with just a few hard negative examples. Experiments on two standard datasets (Flickr30K and COCO) show our approach achieves state-of-the-art dual encoder performance when compared with approaches using a similar amount of data.

8.Representation Compensation Networks for Continual Semantic Segmentation ⬇️

In this work, we study the continual semantic segmentation problem, where the deep neural networks are required to incorporate new classes continually without catastrophic forgetting. We propose to use a structural re-parameterization mechanism, named representation compensation (RC) module, to decouple the representation learning of both old and new knowledge. The RC module consists of two dynamically evolved branches with one frozen and one trainable. Besides, we design a pooled cube knowledge distillation strategy on both spatial and channel dimensions to further enhance the plasticity and stability of the model. We conduct experiments on two challenging continual semantic segmentation scenarios, continual class segmentation and continual domain segmentation. Without any extra computational overhead and parameters during inference, our method outperforms state-of-the-art performance. The code is available at \url{this https URL}.

9.Annotation Efficient Person Re-Identification with Diverse Cluster-Based Pair Selection ⬇️

Person Re-identification (Re-ID) has attracted great attention due to its promising real-world applications. However, in practice, it is always costly to annotate the training data to train a Re-ID model, and it still remains challenging to reduce the annotation cost while maintaining the performance for the Re-ID task. To solve this problem, we propose the Annotation Efficient Person Re-Identification method to select image pairs from an alternative pair set according to the fallibility and diversity of pairs, and train the Re-ID model based on the annotation. Specifically, we design an annotation and training framework to firstly reduce the size of the alternative pair set by clustering all images considering the locality of features, secondly select images pairs from intra-/inter-cluster samples for human to annotate, thirdly re-assign clusters according to the annotation, and finally train the model with the re-assigned clusters. During the pair selection, we seek for valuable pairs according to pairs' fallibility and diversity, which includes an intra-cluster criterion to construct image pairs with the most chaotic samples and the representative samples within clusters, an inter-cluster criterion to construct image pairs between clusters based on the second-order Wasserstein distance, and a diversity criterion for clusterbased pair selection. Combining all criteria above, a greedy strategy is developed to solve the pair selection problem. Finally, the above clustering-selecting-annotating-reassigning-training procedure will be repeated until the annotation budget is reached. Extensive experiments on three widely adopted Re-ID datasets show that we can greatly reduce the annotation cost while achieving better performance compared with state-of-the-art works.

10.Spatial Commonsense Graph for Object Localisation in Partial Scenes ⬇️

We solve object localisation in partial scenes, a new problem of estimating the unknown position of an object (e.g. where is the bag?) given a partial 3D scan of a scene. The proposed solution is based on a novel scene graph model, the Spatial Commonsense Graph (SCG), where objects are the nodes and edges define pairwise distances between them, enriched by concept nodes and relationships from a commonsense knowledge base. This allows SCG to better generalise its spatial inference over unknown 3D scenes. The SCG is used to estimate the unknown position of the target object in two steps: first, we feed the SCG into a novel Proximity Prediction Network, a graph neural network that uses attention to perform distance prediction between the node representing the target object and the nodes representing the observed objects in the SCG; second, we propose a Localisation Module based on circular intersection to estimate the object position using all the predicted pairwise distances in order to be independent of any reference system. We create a new dataset of partially reconstructed scenes to benchmark our method and baselines for object localisation in partial scenes, where our proposed method achieves the best localisation performance.

11.Temporal Context for Robust Maritime Obstacle Detection ⬇️

Robust maritime obstacle detection is essential for fully autonomous unmanned surface vehicles (USVs). The currently widely adopted segmentation-based obstacle detection methods are prone to misclassification of object reflections and sun glitter as obstacles, producing many false positive detections, effectively rendering the methods impractical for USV navigation. However, water-turbulence-induced temporal appearance changes on object reflections are very distinctive from the appearance dynamics of true objects. We harness this property to design WaSR-T, a novel maritime obstacle detection network, that extracts the temporal context from a sequence of recent frames to reduce ambiguity. By learning the local temporal characteristics of object reflection on the water surface, WaSR-T substantially improves obstacle detection accuracy in the presence of reflections and glitter. Compared with existing single-frame methods, WaSR-T reduces the number of false positive detections by 41% overall and by over 53% within the danger zone of the boat, while preserving a high recall, and achieving new state-of-the-art performance on the challenging MODS maritime obstacle detection benchmark.

12.Knowledge-enriched Attention Network with Group-wise Semantic for Visual Storytelling ⬇️

As a technically challenging topic, visual storytelling aims at generating an imaginary and coherent story with narrative multi-sentences from a group of relevant images. Existing methods often generate direct and rigid descriptions of apparent image-based contents, because they are not capable of exploring implicit information beyond images. Hence, these schemes could not capture consistent dependencies from holistic representation, impairing the generation of reasonable and fluent story. To address these problems, a novel knowledge-enriched attention network with group-wise semantic model is proposed. Three main novel components are designed and supported by substantial experiments to reveal practical advantages. First, a knowledge-enriched attention network is designed to extract implicit concepts from external knowledge system, and these concepts are followed by a cascade cross-modal attention mechanism to characterize imaginative and concrete representations. Second, a group-wise semantic module with second-order pooling is developed to explore the globally consistent guidance. Third, a unified one-stage story generation model with encoder-decoder structure is proposed to simultaneously train and infer the knowledge-enriched attention network, group-wise semantic module and multi-modal story generation decoder in an end-to-end fashion. Substantial experiments on the popular Visual Storytelling dataset with both objective and subjective evaluation metrics demonstrate the superior performance of the proposed scheme as compared with other state-of-the-art methods.

13.EyeLoveGAN: Exploiting domain-shifts to boost network learning with cycleGANs ⬇️

This paper presents our contribution to the REFUGE challenge 2020. The challenge consisted of three tasks based on a dataset of retinal images: Segmentation of optic disc and cup, classification of glaucoma, and localization of fovea. We propose employing convolutional neural networks for all three tasks. Segmentation is performed using a U-Net, classification is performed by a pre-trained InceptionV3 network, and fovea detection is performed by employing stacked hour-glass for heatmap prediction. The challenge dataset contains images from three different data sources. To enhance performance, cycleGANs were utilized to create a domain-shift between the data sources. These cycleGANs move images across domains, thus creating artificial images which can be used for training.

14.Domain Generalization via Shuffled Style Assembly for Face Anti-Spoofing ⬇️

With diverse presentation attacks emerging continually, generalizable face anti-spoofing (FAS) has drawn growing attention. Most existing methods implement domain generalization (DG) on the complete representations. However, different image statistics may have unique properties for the FAS tasks. In this work, we separate the complete representation into content and style ones. A novel Shuffled Style Assembly Network (SSAN) is proposed to extract and reassemble different content and style features for a stylized feature space. Then, to obtain a generalized representation, a contrastive learning strategy is developed to emphasize liveness-related style information while suppress the domain-specific one. Finally, the representations of the correct assemblies are used to distinguish between living and spoofing during the inferring. On the other hand, despite the decent performance, there still exists a gap between academia and industry, due to the difference in data quantity and distribution. Thus, a new large-scale benchmark for FAS is built up to further evaluate the performance of algorithms in reality. Both qualitative and quantitative results on existing and proposed benchmarks demonstrate the effectiveness of our methods. The codes will be available at this https URL.

15.TrueType Transformer: Character and Font Style Recognition in Outline Format ⬇️

We propose TrueType Transformer (T3), which can perform character and font style recognition in an outline format. The outline format, such as TrueType, represents each character as a sequence of control points of stroke contours and is frequently used in born-digital documents. T3 is organized by a deep neural network, so-called Transformer. Transformer is originally proposed for sequential data, such as text, and therefore appropriate for handling the outline data. In other words, T3 directly accepts the outline data without converting it into a bitmap image. Consequently, T3 realizes a resolution-independent classification. Moreover, since the locations of the control points represent the fine and local structures of the font style, T3 is suitable for font style classification, where such structures are very important. In this paper, we experimentally show the applicability of T3 in character and font style recognition tasks, while observing how the individual control points contribute to classification results.

16.Non-generative Generalized Zero-shot Learning via Task-correlated Disentanglement and Controllable Samples Synthesis ⬇️

Synthesizing pseudo samples is currently the most effective way to solve the Generalized Zero Shot Learning (GZSL) problem. Most models achieve competitive performance but still suffer from two problems: (1) feature confounding, that task-correlated and task-independent features are confounded in overall representations, which is unreasonable to synthesize reliable pseudo samples; and (2) distribution uncertainty, that massive data is needed when existing models synthesize samples from the uncertain distribution, which causes poor performance in limited samples of seen classes. In this paper, we propose a non-generative model to address these problems correspondingly in two modules: (1) Task-correlated feature disentanglement, to exclude the task-correlated features from task-independent ones by adversarial learning of domain adaption towards reasonable synthesis; and (2) Controllable pseudo sample synthesis, to synthesize edge-pseudo and center-pseudo samples with certain characteristics towards more diversity generated and intuitive transfer. To describe the new scene that is the limit seen class samples in the training process, we further formulate a new ZSL task named the 'Few-shot Seen class and Zero-shot Unseen class learning' (FSZU). Extensive experiments on four benchmarks verify that the proposed method is competitive in the GZSL and the FSZU tasks.

17.Iterative Corresponding Geometry: Fusing Region and Depth for Highly Efficient 3D Tracking of Textureless Objects ⬇️

Tracking objects in 3D space and predicting their 6DoF pose is an essential task in computer vision. State-of-the-art approaches often rely on object texture to tackle this problem. However, while they achieve impressive results, many objects do not contain sufficient texture, violating the main underlying assumption. In the following, we thus propose ICG, a novel probabilistic tracker that fuses region and depth information and only requires the object geometry. Our method deploys correspondence lines and points to iteratively refine the pose. We also implement robust occlusion handling to improve performance in real-world settings. Experiments on the YCB-Video, OPT, and Choi datasets demonstrate that, even for textured objects, our approach outperforms the current state of the art with respect to accuracy and robustness. At the same time, ICG shows fast convergence and outstanding efficiency, requiring only 1.3 ms per frame on a single CPU core. Finally, we analyze the influence of individual components and discuss our performance compared to deep learning-based methods. The source code of our tracker is publicly available.

18.SelfTune: Metrically Scaled Monocular Depth Estimation through Self-Supervised Learning ⬇️

Monocular depth estimation in the wild inherently predicts depth up to an unknown scale. To resolve scale ambiguity issue, we present a learning algorithm that leverages monocular simultaneous localization and mapping (SLAM) with proprioceptive sensors. Such monocular SLAM systems can provide metrically scaled camera poses. Given these metric poses and monocular sequences, we propose a self-supervised learning method for the pre-trained supervised monocular depth networks to enable metrically scaled depth estimation. Our approach is based on a teacher-student formulation which guides our network to predict high-quality depths. We demonstrate that our approach is useful for various applications such as mobile robot navigation and is applicable to diverse environments. Our full system shows improvements over recent self-supervised depth estimation and completion methods on EuRoC, OpenLORIS, and ScanNet datasets.

19.Backbone is All Your Need: A Simplified Architecture for Visual Object Tracking ⬇️

Exploiting a general-purpose neural architecture to replace hand-wired designs or inductive biases has recently drawn extensive interest. However, existing tracking approaches rely on customized sub-modules and need prior knowledge for architecture selection, hindering the tracking development in a more general system. This paper presents a Simplified Tracking architecture (SimTrack) by leveraging a transformer backbone for joint feature extraction and interaction. Unlike existing Siamese trackers, we serialize the input images and concatenate them directly before the one-branch backbone. Feature interaction in the backbone helps to remove well-designed interaction modules and produce a more efficient and effective framework. To reduce the information loss from down-sampling in vision transformers, we further propose a foveal window strategy, providing more diverse input patches with acceptable computational costs. Our SimTrack improves the baseline with 2.5%/2.6% AUC gains on LaSOT/TNL2K and gets results competitive with other specialized tracking algorithms without bells and whistles.

20.StyleBabel: Artistic Style Tagging and Captioning ⬇️

We present StyleBabel, a unique open access dataset of natural language captions and free-form tags describing the artistic style of over 135K digital artworks, collected via a novel participatory method from experts studying at specialist art and design schools. StyleBabel was collected via an iterative method, inspired by `Grounded Theory': a qualitative approach that enables annotation while co-evolving a shared language for fine-grained artistic style attribute description. We demonstrate several downstream tasks for StyleBabel, adapting the recent ALADIN architecture for fine-grained style similarity, to train cross-modal embeddings for: 1) free-form tag generation; 2) natural language description of artistic style; 3) fine-grained text search of style. To do so, we extend ALADIN with recent advances in Visual Transformer (ViT) and cross-modal representation learning, achieving a state of the art accuracy in fine-grained style retrieval.

21.GrainSpace: A Large-scale Dataset for Fine-grained and Domain-adaptive Recognition of Cereal Grains ⬇️

Cereal grains are a vital part of human diets and are important commodities for people's livelihood and international trade. Grain Appearance Inspection (GAI) serves as one of the crucial steps for the determination of grain quality and grain stratification for proper circulation, storage and food processing, etc. GAI is routinely performed manually by qualified inspectors with the aid of some hand tools. Automated GAI has the benefit of greatly assisting inspectors with their jobs but has been limited due to the lack of datasets and clear definitions of the tasks.
In this paper we formulate GAI as three ubiquitous computer vision tasks: fine-grained recognition, domain adaptation and out-of-distribution recognition. We present a large-scale and publicly available cereal grains dataset called GrainSpace. Specifically, we construct three types of device prototypes for data acquisition, and a total of 5.25 million images determined by professional inspectors. The grain samples including wheat, maize and rice are collected from five countries and more than 30 regions. We also develop a comprehensive benchmark based on semi-supervised learning and self-supervised learning techniques. To the best of our knowledge, GrainSpace is the first publicly released dataset for cereal grain inspection.

22.BEAT: A Large-Scale Semantic and Emotional Multi-Modal Dataset for Conversational Gestures Synthesis ⬇️

Achieving realistic, vivid, and human-like synthesized conversational gestures conditioned on multi-modal data is still an unsolved problem, due to the lack of available datasets, models and standard evaluation metrics. To address this, we build Body-Expression-Audio-Text dataset, BEAT, which has i) 76 hours, high-quality, multi-modal data captured from 30 speakers talking with eight different emotions and in four different languages, ii) 32 millions frame-level emotion and semantic relevance annotations.Our statistical analysis on BEAT demonstrates the correlation of conversational gestures with facial expressions, emotions, and semantics, in addition to the known correlation with audio, text, and speaker identity. Qualitative and quantitative experiments demonstrate metrics' validness, ground truth data quality, and baseline's state-of-the-art performance. To the best of our knowledge, BEAT is the largest motion capture dataset for investigating the human gestures, which may contribute to a number of different research fields including controllable gesture synthesis, cross-modality analysis, emotional gesture recognition. The data, code and model will be released for research.

23.Domain Generalisation for Object Detection ⬇️

Domain generalisation aims to promote the learning of domain-invariant features while suppressing domain specific features, so that a model can generalise well on previously unseen target domains. This paper studies domain generalisation in the object detection setting. We propose new terms for handling both the bounding box detector and domain belonging, and incorporate them with consistency regularisation. This allows us to learn a domain agnostic feature representation for object detection, applicable to the problem of domain generalisation. The proposed approach is evaluated using four standard object detection datasets with available domain metadata, namely GWHD, Cityscapes, BDD100K, Sim10K and exhibits consistently superior generalisation performance over baselines.

24.Contrastive Boundary Learning for Point Cloud Segmentation ⬇️

Point cloud segmentation is fundamental in understanding 3D environments. However, current 3D point cloud segmentation methods usually perform poorly on scene boundaries, which degenerates the overall segmentation performance. In this paper, we focus on the segmentation of scene boundaries. Accordingly, we first explore metrics to evaluate the segmentation performance on scene boundaries. To address the unsatisfactory performance on boundaries, we then propose a novel contrastive boundary learning (CBL) framework for point cloud segmentation. Specifically, the proposed CBL enhances feature discrimination between points across boundaries by contrasting their representations with the assistance of scene contexts at multiple scales. By applying CBL on three different baseline methods, we experimentally show that CBL consistently improves different baselines and assists them to achieve compelling performance on boundaries, as well as the overall performance, eg in mIoU. The experimental results demonstrate the effectiveness of our method and the importance of boundaries for 3D point cloud segmentation. Code and model will be made publicly available at this https URL.

25.Real-time Scene Text Detection Based on Global Level and Word Level Features ⬇️

It is an extremely challenging task to detect arbitrary shape text in natural scenes on high accuracy and efficiency. In this paper, we propose a scene text detection framework, namely GWNet, which mainly includes two modules: Global module and RCNN module. Specifically, Global module improves the adaptive performance of the DB (Differentiable Binarization) module by adding k submodule and shift submodule. Two submodules enhance the adaptability of amplifying factor k, accelerate the convergence of models and help to produce more accurate detection results. RCNN module fuses global-level and word-level features. The word-level label is generated by obtaining the minimum axis-aligned rectangle boxes of the shrunk polygon. In the inference period, GWNet only uses global-level features to output simple polygon detections. Experiments on four benchmark datasets, including the MSRA-TD500, Total-Text, ICDAR2015 and CTW-1500, demonstrate that our GWNet outperforms the state-of-the-art detectors. Specifically, with a backbone of ResNet-50, we achieve an F-measure of 88.6% on MSRA- TD500, 87.9% on Total-Text, 89.2% on ICDAR2015 and 87.5% on CTW-1500.

26.A Closer Look at Debiased Temporal Sentence Grounding in Videos: Dataset, Metric, and Approach ⬇️

Temporal Sentence Grounding in Videos (TSGV), which aims to ground a natural language sentence in an untrimmed video, has drawn widespread attention over the past few years. However, recent studies have found that current benchmark datasets may have obvious moment annotation biases, enabling several simple baselines even without training to achieve SOTA performance. In this paper, we take a closer look at existing evaluation protocols, and find both the prevailing dataset and evaluation metrics are the devils that lead to untrustworthy benchmarking. Therefore, we propose to re-organize the two widely-used datasets, making the ground-truth moment distributions different in the training and test splits, i.e., out-of-distribution (OOD) test. Meanwhile, we introduce a new evaluation metric "dR@n,IoU@m" that discounts the basic recall scores to alleviate the inflating evaluation caused by biased datasets. New benchmarking results indicate that our proposed evaluation protocols can better monitor the research progress. Furthermore, we propose a novel causality-based Multi-branch Deconfounding Debiasing (MDD) framework for unbiased moment prediction. Specifically, we design a multi-branch deconfounder to eliminate the effects caused by multiple confounders with causal intervention. In order to help the model better align the semantics between sentence queries and video moments, we enhance the representations during feature encoding. Specifically, for textual information, the query is parsed into several verb-centered phrases to obtain a more fine-grained textual feature. For visual information, the positional information has been decomposed from moment features to enhance representations of moments with diverse locations. Extensive experiments demonstrate that our proposed approach can achieve competitive results among existing SOTA approaches and outperform the base model with great gains.

27.Back to Reality: Weakly-supervised 3D Object Detection with Shape-guided Label Enhancement ⬇️

In this paper, we propose a weakly-supervised approach for 3D object detection, which makes it possible to train strong 3D detector with position-level annotations (i.e. annotations of object centers). In order to remedy the information loss from box annotations to centers, our method, namely Back to Reality (BR), makes use of synthetic 3D shapes to convert the weak labels into fully-annotated virtual scenes as stronger supervision, and in turn utilizes the perfect virtual labels to complement and refine the real labels. Specifically, we first assemble 3D shapes into physically reasonable virtual scenes according to the coarse scene layout extracted from position-level annotations. Then we go back to reality by applying a virtual-to-real domain adaptation method, which refine the weak labels and additionally supervise the training of detector with the virtual scenes. Furthermore, we propose a more challenging benckmark for indoor 3D object detection with more diversity in object sizes to better show the potential of BR. With less than 5% of the labeling labor, we achieve comparable detection performance with some popular fully-supervised approaches on the widely used ScanNet dataset. Code is available at: this https URL

28.Membership Privacy Protection for Image Translation Models via Adversarial Knowledge Distillation ⬇️

Image-to-image translation models are shown to be vulnerable to the Membership Inference Attack (MIA), in which the adversary's goal is to identify whether a sample is used to train the model or not. With daily increasing applications based on image-to-image translation models, it is crucial to protect the privacy of these models against MIAs.
We propose adversarial knowledge distillation (AKD) as a defense method against MIAs for image-to-image translation models. The proposed method protects the privacy of the training samples by improving the generalizability of the model. We conduct experiments on the image-to-image translation models and show that AKD achieves the state-of-the-art utility-privacy tradeoff by reducing the attack performance up to 38.9% compared with the regular training model at the cost of a slight drop in the quality of the generated output images. The experimental results also indicate that the models trained by AKD generalize better than the regular training models. Furthermore, compared with existing defense methods, the results show that at the same privacy protection level, image translation models trained by AKD generate outputs with higher quality; while at the same quality of outputs, AKD enhances the privacy protection over 30%.

29.Transferring Dual Stochastic Graph Convolutional Network for Facial Micro-expression Recognition ⬇️

Micro-expression recognition has drawn increasing attention due to its wide application in lie detection, criminal detection and psychological consultation. To improve the recognition performance of the small micro-expression data, this paper presents a transferring dual stochastic Graph Convolutional Network (TDSGCN) model. We propose a stochastic graph construction method and dual graph convolutional network to extract more discriminative features from the micro-expression images. We use transfer learning to pre-train SGCNs from macro expression data. Optical flow algorithm is also integrated to extract their temporal features. We fuse both spatial and temporal features to improve the recognition performance. To the best of our knowledge, this is the first attempt to utilize the transferring learning and graph convolutional network in micro-expression recognition task. In addition, to handle the class imbalance problem of dataset, we focus on the design of focal loss function. Through extensive evaluation, our proposed method achieves state-of-the-art performance on SAMM and recently released MMEW benchmarks. Our code will be publicly available accompanying this paper.

30.ReF -- Rotation Equivariant Features for Local Feature Matching ⬇️

Sparse local feature matching is pivotal for many computer vision and robotics tasks. To improve their invariance to challenging appearance conditions and viewing angles, and hence their usefulness, existing learning-based methods have primarily focused on data augmentation-based training. In this work, we propose an alternative, complementary approach that centers on inducing bias in the model architecture itself to generate `rotation-specific' features using Steerable E2-CNNs, that are then group-pooled to achieve rotation-invariant local features. We demonstrate that this high performance, rotation-specific coverage from the steerable CNNs can be expanded to all rotation angles by combining it with augmentation-trained standard CNNs which have broader coverage but are often inaccurate, thus creating a state-of-the-art rotation-robust local feature matcher. We benchmark our proposed methods against existing techniques on HPatches and a newly proposed UrbanScenes3D-Air dataset for visual place recognition. Furthermore, we present a detailed analysis of the performance effects of ensembling, robust estimation, network architecture variations, and the use of rotation priors.

31.Crowd Source Scene Change Detection and Local Map Update ⬇️

As scene changes with time map descriptors become outdated, affecting VPS localization accuracy. In this work, we propose an approach to detect structural and texture scene changes to be followed by map update. In our method - map includes 3D points with descriptors generated either via LiDAR or SFM. Common approaches suffer from shortcomings: 1) Direct comparison of the two point-clouds for change detection is slow due to the need to build new point-cloud every time we want to compare; 2) Image based comparison requires to keep the map images adding substantial storage overhead. To circumvent this problems, we propose an approach based on point-clouds descriptors comparison: 1) Based on VPS poses select close query and map images pairs, 2) Registration of query images to map image descriptors, 3) Use segmentation to filter out dynamic or short term temporal changes, 4) Compare the descriptors between corresponding segments.

32.MORE: Multi-Order RElation Mining for Dense Captioning in 3D Scenes ⬇️

3D dense captioning is a recently-proposed novel task, where point clouds contain more geometric information than the 2D counterpart. However, it is also more challenging due to the higher complexity and wider variety of inter-object relations. Existing methods only treat such relations as by-products of object feature learning in graphs without specifically encoding them, which leads to sub-optimal results. In this paper, aiming at improving 3D dense captioning via capturing and utilizing the complex relations in the 3D scene, we propose MORE, a Multi-Order RElation mining model, to support generating more descriptive and comprehensive captions. Technically, our MORE encodes object relations in a progressive manner since complex relations can be deduced from a limited number of basic ones. We first devise a novel Spatial Layout Graph Convolution (SLGC), which semantically encodes several first-order relations as edges of a graph constructed over 3D object proposals. Next, from the resulting graph, we further extract multiple triplets which encapsulate basic first-order relations as the basic unit and construct several Object-centric Triplet Attention Graphs (OTAG) to infer multi-order relations for every target object. The updated node features from OTAG are aggregated and fed into the caption decoder to provide abundant relational cues so that captions including diverse relations with context objects can be generated. Extensive experiments on the Scan2Cap dataset prove the effectiveness of our proposed MORE and its components, and we also outperform the current state-of-the-art method.

33.Online Deep Metric Learning via Mutual Distillation ⬇️

Deep metric learning aims to transform input data into an embedding space, where similar samples are close while dissimilar samples are far apart from each other. In practice, samples of new categories arrive incrementally, which requires the periodical augmentation of the learned model. The fine-tuning on the new categories usually leads to poor performance on the old, which is known as "catastrophic forgetting". Existing solutions either retrain the model from scratch or require the replay of old samples during the training. In this paper, a complete online deep metric learning framework is proposed based on mutual distillation for both one-task and multi-task scenarios. Different from the teacher-student framework, the proposed approach treats the old and new learning tasks with equal importance. No preference over the old or new knowledge is caused. In addition, a novel virtual feature estimation approach is proposed to recover the features assumed to be extracted by the old models. It allows the distillation between the new and the old models without the replay of old training samples or the holding of old models during the training. A comprehensive study shows the superior performance of our approach with the support of different backbones.

34.Hyperspectral Imaging for cherry tomato ⬇️

Cherry tomato (Solanum Lycopersicum) is popular with consumers over the world due to its special flavor. Soluble solids content (SSC) and firmness are two key metrics for evaluating the product qualities. In this work, we develop non-destructive testing techniques for SSC and fruit firmness based on hyperspectral images and a corresponding deep learning regression model. Hyperspectral reflectance images of over 200 tomato fruits are derived with spectrum ranging from 400 to 1000 nm. The acquired hyperspectral images are corrected and the spectral information is extracted. A novel one-dimensional(1D) convolutional ResNet (Con1dResNet) based regression model is prosed and compared with the state of art techniques. Experimental results show that, with a relatively large number of samples our technique is 26.4% better than state of art technique for SSC and 33.7% for firmness. The results of this study indicate the application potential of hyperspectral imaging technique in the SSC and firmness detection, which provides a new option for non-destructive testing of cherry tomato fruit quality in the future.

35.A Screen-Shooting Resilient Document Image Watermarking Scheme using Deep Neural Network ⬇️

With the advent of the screen-reading era, the confidential documents displayed on the screen can be easily captured by a camera without leaving any traces. Thus, this paper proposes a novel screen-shooting resilient watermarking scheme for document image using deep neural network. By applying this scheme, when the watermarked image is displayed on the screen and captured by a camera, the watermark can be still extracted from the captured photographs. Specifically, our scheme is an end-to-end neural network with an encoder to embed watermark and a decoder to extract watermark. During the training process, a distortion layer between encoder and decoder is added to simulate the distortions introduced by screen-shooting process in real scenes, such as camera distortion, shooting distortion, light source distortion. Besides, an embedding strength adjustment strategy is designed to improve the visual quality of the watermarked image with little loss of extraction accuracy. The experimental results show that the scheme has higher robustness and visual quality than other three recent state-of-the-arts. Specially, even if the shooting distances and angles are in extreme, our scheme can also obtain high extraction accuracy.

36.Adaptive Background Matting Using Background Matching ⬇️

Due to the difficulty of solving the matting problem, lots of methods use some kinds of assistance to acquire high quality alpha matte. Green screen matting methods rely on physical equipment. Trimap-based methods take manual interactions as external input. Background-based methods require a pre-captured, static background. The methods are not flexible and convenient enough to use widely. Trimap-free methods are flexible but not stable in complicated video applications. To be stable and flexible in real applications, we propose an adaptive background matting method. The user first captures their videos freely, moving the cameras. Then the user captures the background video afterwards, roughly covering the previous captured regions. We use dynamic background video instead of static background for accurate matting. The proposed method is convenient to use in any scenes as the static camera and background is no more the limitation. To achieve this goal, we use background matching network to find the best-matched background frame by frame from dynamic backgrounds. Then, robust semantic estimation network is used to estimate the coarse alpha matte. Finally, we crop and zoom the target region according to the coarse alpha matte, and estimate the final accurate alpha matte. In experiments, the proposed method is able to perform comparably against the state-of-the-art matting methods.

37.NeRFocus: Neural Radiance Field for 3D Synthetic Defocus ⬇️

Neural radiance fields (NeRF) bring a new wave for 3D interactive experiences. However, as an important part of the immersive experiences, the defocus effects have not been fully explored within NeRF. Some recent NeRF-based methods generate 3D defocus effects in a post-process fashion by utilizing multiplane technology. Still, they are either time-consuming or memory-consuming. This paper proposes a novel thin-lens-imaging-based NeRF framework that can directly render various 3D defocus effects, dubbed NeRFocus. Unlike the pinhole, the thin lens refracts rays of a scene point, so its imaging on the sensor plane is scattered as a circle of confusion (CoC). A direct solution sampling enough rays to approximate this process is computationally expensive. Instead, we propose to inverse the thin lens imaging to explicitly model the beam path for each point on the sensor plane and generalize this paradigm to the beam path of each pixel, then use the frustum-based volume rendering to render each pixel's beam path. We further design an efficient probabilistic training (p-training) strategy to simplify the training process vastly. Extensive experiments demonstrate that our NeRFocus can achieve various 3D defocus effects with adjustable camera pose, focus distance, and aperture size. Existing NeRF can be regarded as our special case by setting aperture size as zero to render large depth-of-field images. Despite such merits, NeRFocus does not sacrifice NeRF's original performance (e.g., training and inference time, parameter consumption, rendering quality), which implies its great potential for broader application and further improvement.

38.Suspected Object Matters: Rethinking Model's Prediction for One-stage Visual Grounding ⬇️

Recently, one-stage visual grounders attract high attention due to the comparable accuracy but significantly higher efficiency than two-stage grounders. However, inter-object relation modeling has not been well studied for one-stage grounders. Inter-object relationship modeling, though important, is not necessarily performed among all the objects within the image, as only a part of them are related to the text query and may confuse the model. We call these objects "suspected objects". However, exploring relationships among these suspected objects in the one-stage visual grounding paradigm is non-trivial due to two core problems: (1) no object proposals are available as the basis on which to select suspected objects and perform relationship modeling; (2) compared with those irrelevant to the text query, suspected objects are more confusing, as they may share similar semantics, be entangled with certain relationships, etc, and thereby more easily mislead the model's prediction. To address the above issues, this paper proposes a Suspected Object Graph (SOG) approach to encourage the correct referred object selection among the suspected ones in the one-stage visual grounding. Suspected objects are dynamically selected from a learned activation map as nodes to adapt to the current discrimination ability of the model during training. Afterward, on top of the suspected objects, a Keyword-aware Node Representation module (KNR) and an Exploration by Random Connection strategy (ERC) are concurrently proposed within the SOG to help the model rethink its initial prediction. Extensive ablation studies and comparison with state-of-the-art approaches on prevalent visual grounding benchmarks demonstrate the effectiveness of our proposed method.

39.Knowledge Distillation as Efficient Pre-training: Faster Convergence, Higher Data-efficiency, and Better Transferability ⬇️

Large-scale pre-training has been proven to be crucial for various computer vision tasks. However, with the increase of pre-training data amount, model architecture amount, and the private/inaccessible data, it is not very efficient or possible to pre-train all the model architectures on large-scale datasets. In this work, we investigate an alternative strategy for pre-training, namely Knowledge Distillation as Efficient Pre-training (KDEP), aiming to efficiently transfer the learned feature representation from existing pre-trained models to new student models for future downstream tasks. We observe that existing Knowledge Distillation (KD) methods are unsuitable towards pre-training since they normally distill the logits that are going to be discarded when transferred to downstream tasks. To resolve this problem, we propose a feature-based KD method with non-parametric feature dimension aligning. Notably, our method performs comparably with supervised pre-training counterparts in 3 downstream tasks and 9 downstream datasets requiring 10x less data and 5x less pre-training time. Code is available at this https URL.

40.Towards Open-Set Text Recognition via Label-to-Prototype Learning ⬇️

Scene text recognition is a popular topic and can benefit various tasks. Although many methods have been proposed for the close-set text recognition challenges, they cannot be directly applied to open-set scenarios, where the evaluation set contains novel characters not appearing in the training set. Conventional methods require collecting new data and retraining the model to handle these novel characters, which is an expensive and tedious process. In this paper, we propose a label-to-prototype learning framework to handle novel characters without retraining the model. In the proposed framework, novel characters are effectively mapped to their corresponding prototypes with a label-to-prototype learning module. This module is trained on characters with seen labels and can be easily generalized to novel characters. Additionally, feature-level rectification is conducted via topology-preserving transformation, resulting in better alignments between visual features and constructed prototypes while having a reasonably small impact on model speed. A lot of experiments show that our method achieves promising performance on a variety of zero-shot, close-set, and open-set text recognition datasets.

41.An Audio-Visual Attention Based Multimodal Network for Fake Talking Face Videos Detection ⬇️

DeepFake based digital facial forgery is threatening the public media security, especially when lip manipulation has been used in talking face generation, the difficulty of fake video detection is further improved. By only changing lip shape to match the given speech, the facial features of identity is hard to be discriminated in such fake talking face videos. Together with the lack of attention on audio stream as the prior knowledge, the detection failure of fake talking face generation also becomes inevitable. Inspired by the decision-making mechanism of human multisensory perception system, which enables the auditory information to enhance post-sensory visual evidence for informed decisions output, in this study, a fake talking face detection framework FTFDNet is proposed by incorporating audio and visual representation to achieve more accurate fake talking face videos detection. Furthermore, an audio-visual attention mechanism (AVAM) is proposed to discover more informative features, which can be seamlessly integrated into any audio-visual CNN architectures by modularization. With the additional AVAM, the proposed FTFDNet is able to achieve a better detection performance on the established dataset (FTFDD). The evaluation of the proposed work has shown an excellent performance on the detection of fake talking face videos, which is able to arrive at a detection rate above 97%.

42.MVP: Multimodality-guided Visual Pre-training ⬇️

Recently, masked image modeling (MIM) has become a promising direction for visual pre-training. In the context of vision transformers, MIM learns effective visual representation by aligning the token-level features with a pre-defined space (e.g., BEIT used a d-VAE trained on a large image corpus as the tokenizer). In this paper, we go one step further by introducing guidance from other modalities and validating that such additional knowledge leads to impressive gains for visual pre-training. The proposed approach is named Multimodality-guided Visual Pre-training (MVP), in which we replace the tokenizer with the vision branch of CLIP, a vision-language model pre-trained on 400 million image-text pairs. We demonstrate the effectiveness of MVP by performing standard experiments, i.e., pre-training the ViT models on ImageNet and fine-tuning them on a series of downstream visual recognition tasks. In particular, pre-training ViT-Base/16 for 300 epochs, MVP reports a 52.4% mIoU on ADE20K, surpassing BEIT (the baseline and previous state-of-the-art) with an impressive margin of 6.8%.

43.Zero-Shot Action Recognition with Transformer-based Video Semantic Embedding ⬇️

While video action recognition has been an active area of research for several years, zero-shot action recognition has only recently started gaining traction. However, there is a lack of a formal definition for the zero-shot learning paradigm leading to uncertainty about classes that can be considered as previously unseen. In this work, we take a new comprehensive look at the inductive zero-shot action recognition problem from a realistic standpoint. Specifically, we advocate for a concrete formulation for zero-shot action recognition that avoids an exact overlap between the training and testing classes and also limits the intra-class variance; and propose a novel end-to-end trained transformer model which is capable of capturing long range spatiotemporal dependencies efficiently, contrary to existing approaches which use 3D-CNNs. The proposed approach outperforms the existing state-of-the-art algorithms in many settings on all benchmark datasets by a wide margin.

44.Practical Evaluation of Adversarial Robustness via Adaptive Auto Attack ⬇️

Defense models against adversarial attacks have grown significantly, but the lack of practical evaluation methods has hindered progress. Evaluation can be defined as looking for defense models' lower bound of robustness given a budget number of iterations and a test dataset. A practical evaluation method should be convenient (i.e., parameter-free), efficient (i.e., fewer iterations) and reliable (i.e., approaching the lower bound of robustness). Towards this target, we propose a parameter-free Adaptive Auto Attack (A$^3$) evaluation method which addresses the efficiency and reliability in a test-time-training fashion. Specifically, by observing that adversarial examples to a specific defense model follow some regularities in their starting points, we design an Adaptive Direction Initialization strategy to speed up the evaluation. Furthermore, to approach the lower bound of robustness under the budget number of iterations, we propose an online statistics-based discarding strategy that automatically identifies and abandons hard-to-attack images. Extensive experiments demonstrate the effectiveness of our A$^3$. Particularly, we apply A$^3$ to nearly 50 widely-used defense models. By consuming much fewer iterations than existing methods, i.e., $1/10$ on average (10$\times$ speed up), we achieve lower robust accuracy in all cases. Notably, we won $\textbf{first place}$ out of 1681 teams in CVPR 2021 White-box Adversarial Attacks on Defense Models competitions with this method. Code is available at: $\href{this https URL}{this https URL_auto_attack}$

45.Frequency-driven Imperceptible Adversarial Attack on Semantic Similarity ⬇️

Current adversarial attack research reveals the vulnerability of learning-based classifiers against carefully crafted perturbations. However, most existing attack methods have inherent limitations in cross-dataset generalization as they rely on a classification layer with a closed set of categories. Furthermore, the perturbations generated by these methods may appear in regions easily perceptible to the human visual system (HVS). To circumvent the former problem, we propose a novel algorithm that attacks semantic similarity on feature representations. In this way, we are able to fool classifiers without limiting attacks to a specific dataset. For imperceptibility, we introduce the low-frequency constraint to limit perturbations within high-frequency components, ensuring perceptual similarity between adversarial examples and originals. Extensive experiments on three datasets (CIFAR-10, CIFAR-100, and ImageNet-1K) and three public online platforms indicate that our attack can yield misleading and transferable adversarial examples across architectures and datasets. Additionally, visualization results and quantitative performance (in terms of four different metrics) show that the proposed algorithm generates more imperceptible perturbations than the state-of-the-art methods. Code is made available at.

46.Intention-aware Feature Propagation Network for Interactive Segmentation ⬇️

We aim to tackle the problem of point-based interactive segmentation, in which two key challenges are to infer user's intention correctly and to propagate the user-provided annotations to unlabeled regions efficiently. To address those challenges, we propose a novel intention-aware feature propagation strategy that performs explicit user intention estimation and learns an efficient click-augmented feature representation for high-resolution foreground segmentation. Specifically, we develop a coarse-to-fine sparse propagation network for each interactive segmentation step, which consists of a coarse-level network for more effective tracking of user's interest, and a fine-level network for zooming to the target object and performing fine-level segmentation. Moreover, we design a new sparse graph network module for both levels to enable efficient long-range propagation of click information. Extensive experiments show that our method surpasses the previous state-of-the-art methods on all popular benchmarks, demonstrating its efficacy.

47.Cross-modal Map Learning for Vision and Language Navigation ⬇️

We consider the problem of Vision-and-Language Navigation (VLN). The majority of current methods for VLN are trained end-to-end using either unstructured memory such as LSTM, or using cross-modal attention over the egocentric observations of the agent. In contrast to other works, our key insight is that the association between language and vision is stronger when it occurs in explicit spatial representations. In this work, we propose a cross-modal map learning model for vision-and-language navigation that first learns to predict the top-down semantics on an egocentric map for both observed and unobserved regions, and then predicts a path towards the goal as a set of waypoints. In both cases, the prediction is informed by the language through cross-modal attention mechanisms. We experimentally test the basic hypothesis that language-driven navigation can be solved given a map, and then show competitive results on the full VLN-CE benchmark.

48.Manifold Modeling in Quotient Space: Learning An Invariant Mapping with Decodability of Image Patches ⬇️

This study proposes a framework for manifold learning of image patches using the concept of equivalence classes: manifold modeling in quotient space (MMQS). In MMQS, we do not consider a set of local patches of the image as it is, but rather the set of their canonical patches obtained by introducing the concept of equivalence classes and performing manifold learning on their canonical patches. Canonical patches represent equivalence classes, and their auto-encoder constructs a manifold in the quotient space. Based on this framework, we produce a novel manifold-based image model by introducing rotation-flip-equivalence relations. In addition, we formulate an image reconstruction problem by fitting the proposed image model to a corrupted observed image and derive an algorithm to solve it. Our experiments show that the proposed image model is effective for various self-supervised image reconstruction tasks, such as image inpainting, deblurring, super-resolution, and denoising.

49.DEER: Detection-agnostic End-to-End Recognizer for Scene Text Spotting ⬇️

Recent end-to-end scene text spotters have achieved great improvement in recognizing arbitrary-shaped text instances. Common approaches for text spotting use region of interest pooling or segmentation masks to restrict features to single text instances. However, this makes it hard for the recognizer to decode correct sequences when the detection is not accurate i.e. one or more characters are cropped out. Considering that it is hard to accurately decide word boundaries with only the detector, we propose a novel Detection-agnostic End-to-End Recognizer, DEER, framework. The proposed method reduces the tight dependency between detection and recognition modules by bridging them with a single reference point for each text instance, instead of using detected regions. The proposed method allows the decoder to recognize the texts that are indicated by the reference point, with features from the whole image. Since only a single point is required to recognize the text, the proposed method enables text spotting without an arbitrarily-shaped detector or bounding polygon annotations. Experimental results present that the proposed method achieves competitive results on regular and arbitrarily-shaped text spotting benchmarks. Further analysis shows that DEER is robust to the detection errors. The code and dataset will be publicly available.

50.MetAug: Contrastive Learning via Meta Feature Augmentation ⬇️

What matters for contrastive learning? We argue that contrastive learning heavily relies on informative features, or "hard" (positive or negative) features. Early works include more informative features by applying complex data augmentations and large batch size or memory bank, and recent works design elaborate sampling approaches to explore informative features. The key challenge toward exploring such features is that the source multi-view data is generated by applying random data augmentations, making it infeasible to always add useful information in the augmented data. Consequently, the informativeness of features learned from such augmented data is limited. In response, we propose to directly augment the features in latent space, thereby learning discriminative representations without a large amount of input data. We perform a meta learning technique to build the augmentation generator that updates its network parameters by considering the performance of the encoder. However, insufficient input data may lead the encoder to learn collapsed features and therefore malfunction the augmentation generator. A new margin-injected regularization is further added in the objective function to avoid the encoder learning a degenerate mapping. To contrast all features in one gradient back-propagation step, we adopt the proposed optimization-driven unified contrastive loss instead of the conventional contrastive loss. Empirically, our method achieves state-of-the-art results on several benchmark datasets.

51.Semi-supervision semantic segmentation with uncertainty-guided self cross supervision ⬇️

As a powerful way of realizing semi-supervised segmentation, the cross supervision method learns cross consistency based on independent ensemble models using abundant unlabeled images. However, the wrong pseudo labeling information generated by cross supervision would confuse the training process and negatively affect the effectiveness of the segmentation model. Besides, the training process of ensemble models in such methods also multiplies the cost of computation resources and decreases the training efficiency. To solve these problems, we propose a novel cross supervision method, namely uncertainty-guided self cross supervision (USCS). In addition to ensemble models, we first design a multi-input multi-output (MIMO) segmentation model which can generate multiple outputs with shared model and consequently impose consistency over the outputs, saving the cost on parameters and calculations. On the other hand, we employ uncertainty as guided information to encourage the model to focus on the high confident regions of pseudo labels and mitigate the effects of wrong pseudo labeling in self cross supervision, improving the performance of the segmentation model. Extensive experiments show that our method achieves state-of-the-art performance while saving 40.5% and 49.1% cost on parameters and calculations.

52.OpenTAL: Towards Open Set Temporal Action Localization ⬇️

Temporal Action Localization (TAL) has experienced remarkable success under the supervised learning paradigm. However, existing TAL methods are rooted in the closed set assumption, which cannot handle the inevitable unknown actions in open-world scenarios. In this paper, we, for the first time, step toward the Open Set TAL (OSTAL) problem and propose a general framework OpenTAL based on Evidential Deep Learning (EDL). Specifically, the OpenTAL consists of uncertainty-aware action classification, actionness prediction, and temporal location regression. With the proposed importance-balanced EDL method, classification uncertainty is learned by collecting categorical evidence majorly from important samples. To distinguish the unknown actions from background video frames, the actionness is learned by the positive-unlabeled learning. The classification uncertainty is further calibrated by leveraging the guidance from the temporal localization quality. The OpenTAL is general to enable existing TAL models for open set scenarios, and experimental results on THUMOS14 and ActivityNet1.3 benchmarks show the effectiveness of our method. The code and pre-trained models are released at this https URL.

53.Improving Neural ODEs via Knowledge Distillation ⬇️

Neural Ordinary Differential Equations (Neural ODEs) construct the continuous dynamics of hidden units using ordinary differential equations specified by a neural network, demonstrating promising results on many tasks. However, Neural ODEs still do not perform well on image recognition tasks. The possible reason is that the one-hot encoding vector commonly used in Neural ODEs can not provide enough supervised information. We propose a new training based on knowledge distillation to construct more powerful and robust Neural ODEs fitting image recognition tasks. Specially, we model the training of Neural ODEs into a teacher-student learning process, in which we propose ResNets as the teacher model to provide richer supervised information. The experimental results show that the new training manner can improve the classification accuracy of Neural ODEs by 24% on CIFAR10 and 5% on SVHN. In addition, we also quantitatively discuss the effect of both knowledge distillation and time horizon in Neural ODEs on robustness against adversarial examples. The experimental analysis concludes that introducing the knowledge distillation and increasing the time horizon can improve the robustness of Neural ODEs against adversarial examples.

54.NLX-GPT: A Model for Natural Language Explanations in Vision and Vision-Language Tasks ⬇️

Natural language explanation (NLE) models aim at explaining the decision-making process of a black box system via generating natural language sentences which are human-friendly, high-level and fine-grained. Current NLE models explain the decision-making process of a vision or vision-language model (a.k.a., task model), e.g., a VQA model, via a language model (a.k.a., explanation model), e.g., GPT. Other than the additional memory resources and inference time required by the task model, the task and explanation models are completely independent, which disassociates the explanation from the reasoning process made to predict the answer. We introduce NLX-GPT, a general, compact and faithful language model that can simultaneously predict an answer and explain it. We first conduct pre-training on large scale data of image-caption pairs for general understanding of images, and then formulate the answer as a text prediction task along with the explanation. Without region proposals nor a task model, our resulting overall framework attains better evaluation scores, contains much less parameters and is 15$\times$ faster than the current SoA model. We then address the problem of evaluating the explanations which can be in many times generic, data-biased and can come in several forms. We therefore design 2 new evaluation measures: (1) explain-predict and (2) retrieval-based attack, a self-evaluation framework that requires no labels. Code is at: this https URL.

55.SynWoodScape: Synthetic Surround-view Fisheye Camera Dataset for Autonomous Driving ⬇️

Surround-view cameras are a primary sensor for automated driving, used for near field perception. It is one of the most commonly used sensors in commercial vehicles. Four fisheye cameras with a 190° field of view cover the 360° around the vehicle. Due to its high radial distortion, the standard algorithms do not extend easily. Previously, we released the first public fisheye surround-view dataset named WoodScape. In this work, we release a synthetic version of the surround-view dataset, covering many of its weaknesses and extending it. Firstly, it is not possible to obtain ground truth for pixel-wise optical flow and depth. Secondly, WoodScape did not have all four cameras simultaneously in order to sample diverse frames. However, this means that multi-camera algorithms cannot be designed, which is enabled in the new dataset. We implemented surround-view fisheye geometric projections in CARLA Simulator matching WoodScape's configuration and created SynWoodScape. We release 80k images from the synthetic dataset with annotations for 10+ tasks. We also release the baseline code and supporting scripts.

56.Optical Flow Training under Limited Label Budget via Active Learning ⬇️

Supervised training of optical flow predictors generally yields better accuracy than unsupervised training. However, the improved performance comes at an often high annotation cost. Semi-supervised training trades off accuracy against annotation cost. We use a simple yet effective semi-supervised training method to show that even a small fraction of labels can improve flow accuracy by a significant margin over unsupervised training. In addition, we propose active learning methods based on simple heuristics to further reduce the number of labels required to achieve the same target accuracy. Our experiments on both synthetic and real optical flow datasets show that our semi-supervised networks generally need around 50% of the labels to achieve close to full-label accuracy, and only around 20% with active learning on Sintel. We also analyze and show insights on the factors that may influence our active learning performance. Code will be made available soon.

57.Evaluating Proposed Fairness Models for Face Recognition Algorithms ⬇️

The development of face recognition algorithms by academic and commercial organizations is growing rapidly due to the onset of deep learning and the widespread availability of training data. Though tests of face recognition algorithm performance indicate yearly performance gains, error rates for many of these systems differ based on the demographic composition of the test set. These "demographic differentials" in algorithm performance can contribute to unequal or unfair outcomes for certain groups of people, raising concerns with increased worldwide adoption of face recognition systems. Consequently, regulatory bodies in both the United States and Europe have proposed new rules requiring audits of biometric systems for "discriminatory impacts" (European Union Artificial Intelligence Act) and "fairness" (U.S. Federal Trade Commission). However, no standard for measuring fairness in biometric systems yet exists. This paper characterizes two proposed measures of face recognition algorithm fairness (fairness measures) from scientists in the U.S. and Europe. We find that both proposed methods are challenging to interpret when applied to disaggregated face recognition error rates as they are commonly experienced in practice. To address this, we propose a set of interpretability criteria, termed the Functional Fairness Measure Criteria (FFMC), that outlines a set of properties desirable in a face recognition algorithm fairness measure. We further develop a new fairness measure, the Gini Aggregation Rate for Biometric Equitability (GARBE), and show how, in conjunction with the Pareto optimization, this measure can be used to select among alternative algorithms based on the accuracy/fairness trade-space. Finally, we have open-sourced our dataset of machine-readable, demographically disaggregated error rates. We believe this is currently the largest open-source dataset of its kind.

58.Adaptive Trajectory Prediction via Transferable GNN ⬇️

Pedestrian trajectory prediction is an essential component in a wide range of AI applications such as autonomous driving and robotics. Existing methods usually assume the training and testing motions follow the same pattern while ignoring the potential distribution differences (e.g., shopping mall and street). This issue results in inevitable performance decrease. To address this issue, we propose a novel Transferable Graph Neural Network (T-GNN) framework, which jointly conducts trajectory prediction as well as domain alignment in a unified framework. Specifically, a domain invariant GNN is proposed to explore the structural motion knowledge where the domain specific knowledge is reduced. Moreover, an attention-based adaptive knowledge learning module is further proposed to explore fine-grained individual-level feature representation for knowledge transfer. By this way, disparities across different trajectory domains will be better alleviated. More challenging while practical trajectory prediction experiments are designed, and the experimental results verify the superior performance of our proposed model. To the best of our knowledge, our work is the pioneer which fills the gap in benchmarks and techniques for practical pedestrian trajectory prediction across different domains.

59.Dynamic Instance Domain Adaptation ⬇️

Most existing studies on unsupervised domain adaptation (UDA) assume that each domain's training samples come with domain labels (e.g., painting, photo). Samples from each domain are assumed to follow the same distribution and the domain labels are exploited to learn domain-invariant features via feature alignment. However, such an assumption often does not hold true -- there often exist numerous finer-grained domains (e.g., dozens of modern painting styles have been developed, each differing dramatically from those of the classic styles). Therefore, forcing feature distribution alignment across each artificially-defined and coarse-grained domain can be ineffective. In this paper, we address both single-source and multi-source UDA from a completely different perspective, which is to view each instance as a fine domain. Feature alignment across domains is thus redundant. Instead, we propose to perform dynamic instance domain adaptation (DIDA). Concretely, a dynamic neural network with adaptive convolutional kernels is developed to generate instance-adaptive residuals to adapt domain-agnostic deep features to each individual instance. This enables a shared classifier to be applied to both source and target domain data without relying on any domain annotation. Further, instead of imposing intricate feature alignment losses, we adopt a simple semi-supervised learning paradigm using only a cross-entropy loss for both labeled source and pseudo labeled target data. Our model, dubbed DIDA-Net, achieves state-of-the-art performance on several commonly used single-source and multi-source UDA datasets including Digits, Office-Home, DomainNet, Digit-Five, and PACS.

60.Resource-Efficient Invariant Networks: Exponential Gains by Unrolled Optimization ⬇️

Achieving invariance to nuisance transformations is a fundamental challenge in the construction of robust and reliable vision systems. Existing approaches to invariance scale exponentially with the dimension of the family of transformations, making them unable to cope with natural variabilities in visual data such as changes in pose and perspective. We identify a common limitation of these approaches--they rely on sampling to traverse the high-dimensional space of transformations--and propose a new computational primitive for building invariant networks based instead on optimization, which in many scenarios provides a provably more efficient method for high-dimensional exploration than sampling. We provide empirical and theoretical corroboration of the efficiency gains and soundness of our proposed method, and demonstrate its utility in constructing an efficient invariant network for a simple hierarchical object detection task when combined with unrolled optimization. Code for our networks and experiments is available at this https URL.

61.Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time ⬇️

The conventional recipe for maximizing model accuracy is to (1) train multiple models with various hyperparameters and (2) pick the individual model which performs best on a held-out validation set, discarding the remainder. In this paper, we revisit the second step of this procedure in the context of fine-tuning large pre-trained models, where fine-tuned models often appear to lie in a single low error basin. We show that averaging the weights of multiple models fine-tuned with different hyperparameter configurations often improves accuracy and robustness. Unlike a conventional ensemble, we may average many models without incurring any additional inference or memory costs -- we call the results "model soups." When fine-tuning large pre-trained models such as CLIP, ALIGN, and a ViT-G pre-trained on JFT, our soup recipe provides significant improvements over the best model in a hyperparameter sweep on ImageNet. As a highlight, the resulting ViT-G model attains 90.94% top-1 accuracy on ImageNet, a new state of the art. Furthermore, we show that the model soup approach extends to multiple image classification and natural language processing tasks, improves out-of-distribution performance, and improves zero-shot performance on new downstream tasks. Finally, we analytically relate the performance similarity of weight-averaging and logit-ensembling to flatness of the loss and confidence of the predictions, and validate this relation empirically.

62.Two-stream Hierarchical Similarity Reasoning for Image-text Matching ⬇️

Reasoning-based approaches have demonstrated their powerful ability for the task of image-text matching. In this work, two issues are addressed for image-text matching. First, for reasoning processing, conventional approaches have no ability to find and use multi-level hierarchical similarity information. To solve this problem, a hierarchical similarity reasoning module is proposed to automatically extract context information, which is then co-exploited with local interaction information for efficient reasoning. Second, previous approaches only consider learning single-stream similarity alignment (i.e., image-to-text level or text-to-image level), which is inadequate to fully use similarity information for image-text matching. To address this issue, a two-stream architecture is developed to decompose image-text matching into image-to-text level and text-to-image level similarity computation. These two issues are investigated by a unifying framework that is trained in an end-to-end manner, namely two-stream hierarchical similarity reasoning network. The extensive experiments performed on the two benchmark datasets of MSCOCO and Flickr30K show the superiority of the proposed approach as compared to existing state-of-the-art methods.

63.Cluttered Food Grasping with Adaptive Fingers and Synthetic-Data Trained Object Detection ⬇️

The food packaging industry handles an immense variety of food products with wide-ranging shapes and sizes, even within one kind of food. Menus are also diverse and change frequently, making automation of pick-and-place difficult. A popular approach to bin-picking is to first identify each piece of food in the tray by using an instance segmentation method. However, human annotations to train these methods are unreliable and error-prone since foods are packed close together with unclear boundaries and visual similarity making separation of pieces difficult. To address this problem, we propose a method that trains purely on synthetic data and successfully transfers to the real world using sim2real methods by creating datasets of filled food trays using high-quality 3d models of real pieces of food for the training instance segmentation models. Another concern is that foods are easily damaged during grasping. We address this by introducing two additional methods -- a novel adaptive finger mechanism to passively retract when a collision occurs, and a method to filter grasps that are likely to cause damage to neighbouring pieces of food during a grasp. We demonstrate the effectiveness of the proposed method on several kinds of real foods.

64.A Tree-Structured Multi-Task Model Recommender ⬇️

Tree-structured multi-task architectures have been employed to jointly tackle multiple vision tasks in the context of multi-task learning (MTL). The major challenge is to determine where to branch out for each task given a backbone model to optimize for both task accuracy and computation efficiency. To address the challenge, this paper proposes a recommender that, given a set of tasks and a convolutional neural network-based backbone model, automatically suggests tree-structured multi-task architectures that could achieve a high task performance while meeting a user-specified computation budget without performing model training. Extensive evaluations on popular MTL benchmarks show that the recommended architectures could achieve competitive task accuracy and computation efficiency compared with state-of-the-art MTL methods.

65.The Transitive Information Theory and its Application to Deep Generative Models ⬇️

Paradoxically, a Variational Autoencoder (VAE) could be pushed in two opposite directions, utilizing powerful decoder model for generating realistic images but collapsing the learned representation, or increasing regularization coefficient for disentangling representation but ultimately generating blurry examples. Existing methods narrow the issues to the rate-distortion trade-off between compression and reconstruction. We argue that a good reconstruction model does learn high capacity latents that encode more details, however, its use is hindered by two major issues: the prior is random noise which is completely detached from the posterior and allow no controllability in the generation; mean-field variational inference doesn't enforce hierarchy structure which makes the task of recombining those units into plausible novel output infeasible. As a result, we develop a system that learns a hierarchy of disentangled representation together with a mechanism for recombining the learned representation for generalization. This is achieved by introducing a minimal amount of inductive bias to learn controllable prior for the VAE. The idea is supported by here developed transitive information theory, that is, the mutual information between two target variables could alternately be maximized through the mutual information to the third variable, thus bypassing the rate-distortion bottleneck in VAE design. In particular, we show that our model, named SemafoVAE (inspired by the similar concept in computer science), could generate high-quality examples in a controllable manner, perform smooth traversals of the disentangled factors and intervention at a different level of representation hierarchy.

66.UNeXt: MLP-based Rapid Medical Image Segmentation Network ⬇️

UNet and its latest extensions like TransUNet have been the leading medical image segmentation methods in recent years. However, these networks cannot be effectively adopted for rapid image segmentation in point-of-care applications as they are parameter-heavy, computationally complex and slow to use. To this end, we propose UNeXt which is a Convolutional multilayer perceptron (MLP) based network for image segmentation. We design UNeXt in an effective way with an early convolutional stage and a MLP stage in the latent stage. We propose a tokenized MLP block where we efficiently tokenize and project the convolutional features and use MLPs to model the representation. To further boost the performance, we propose shifting the channels of the inputs while feeding in to MLPs so as to focus on learning local dependencies. Using tokenized MLPs in latent space reduces the number of parameters and computational complexity while being able to result in a better representation to help segmentation. The network also consists of skip connections between various levels of encoder and decoder. We test UNeXt on multiple medical image segmentation datasets and show that we reduce the number of parameters by 72x, decrease the computational complexity by 68x, and improve the inference speed by 10x while also obtaining better segmentation performance over the state-of-the-art medical image segmentation architectures. Code is available at this https URL

67.Metastatic Cancer Outcome Prediction with Injective Multiple Instance Pooling ⬇️

Cancer stage is a large determinant of patient prognosis and management in many cancer types, and is often assessed using medical imaging modalities, such as CT and MRI. These medical images contain rich information that can be explored to stratify patients within each stage group to further improve prognostic algorithms. Although the majority of cancer deaths result from metastatic and multifocal disease, building imaging biomarkers for patients with multiple tumors has been a challenging task due to the lack of annotated datasets and standard study framework. In this paper, we process two public datasets to set up a benchmark cohort of 341 patient in total for studying outcome prediction of multifocal metastatic cancer. We identify the lack of expressiveness in common multiple instance classification networks and propose two injective multiple instance pooling functions that are better suited to outcome prediction. Our results show that multiple instance learning with injective pooling functions can achieve state-of-the-art performance in the non-small-cell lung cancer CT and head and neck CT outcome prediction benchmarking tasks. We will release the processed multifocal datasets, our code and the intermediate files i.e. extracted radiomic features to support further transparent and reproducible research.

68.Neural Data-Dependent Transform for Learned Image Compression ⬇️

Learned image compression has achieved great success due to its excellent modeling capacity, but seldom further considers the Rate-Distortion Optimization (RDO) of each input image. To explore this potential in the learned codec, we make the first attempt to build a neural data-dependent transform and introduce a continuous online mode decision mechanism to jointly optimize the coding efficiency for each individual image. Specifically, apart from the image content stream, we employ an additional model stream to generate the transform parameters at the decoder side. The presence of a model stream enables our model to learn more abstract neural-syntax, which helps cluster the latent representations of images more compactly. Beyond the transform stage, we also adopt neural-syntax based post-processing for the scenarios that require higher quality reconstructions regardless of extra decoding overhead. Moreover, the involvement of the model stream further makes it possible to optimize both the representation and the decoder in an online way, i.e. RDO at the testing time. It is equivalent to a continuous online mode decision, like coding modes in the traditional codecs, to improve the coding efficiency based on the individual input image. The experimental results show the effectiveness of the proposed neural-syntax design and the continuous online mode decision mechanism, demonstrating the superiority of our method in coding efficiency compared to the latest conventional standard Versatile Video Coding (VVC) and other state-of-the-art learning-based methods.

69.Learning the Degradation Distribution for Blind Image Super-Resolution ⬇️

Synthetic high-resolution (HR) & low-resolution (LR) pairs are widely used in existing super-resolution (SR) methods. To avoid the domain gap between synthetic and test images, most previous methods try to adaptively learn the synthesizing (degrading) process via a deterministic model. However, some degradations in real scenarios are stochastic and cannot be determined by the content of the image. These deterministic models may fail to model the random factors and content-independent parts of degradations, which will limit the performance of the following SR models. In this paper, we propose a probabilistic degradation model (PDM), which studies the degradation $\mathbf{D}$ as a random variable, and learns its distribution by modeling the mapping from a priori random variable $\mathbf{z}$ to $\mathbf{D}$. Compared with previous deterministic degradation models, PDM could model more diverse degradations and generate HR-LR pairs that may better cover the various degradations of test images, and thus prevent the SR model from over-fitting to specific ones. Extensive experiments have demonstrated that our degradation model can help the SR model achieve better performance on different datasets. The source codes are released at \url{[email protected]:greatlog/UnpairedSR.git}.

70.Sharing Generative Models Instead of Private Data: A Simulation Study on Mammography Patch Classification ⬇️

Early detection of breast cancer in mammography screening via deep-learning based computer-aided detection systems shows promising potential in improving the curability and mortality rates of breast cancer. However, many clinical centres are restricted in the amount and heterogeneity of available data to train such models to (i) achieve promising performance and to (ii) generalise well across acquisition protocols and domains. As sharing data between centres is restricted due to patient privacy concerns, we propose a potential solution: sharing trained generative models between centres as substitute for real patient data. In this work, we use three well known mammography datasets to simulate three different centres, where one centre receives the trained generator of Generative Adversarial Networks (GANs) from the two remaining centres in order to augment the size and heterogeneity of its training dataset. We evaluate the utility of this approach on mammography patch classification on the test set of the GAN-receiving centre using two different classification models, (a) a convolutional neural network and (b) a transformer neural network. Our experiments demonstrate that shared GANs notably increase the performance of both transformer and convolutional classification models and highlight this approach as a viable alternative to inter-centre data sharing.

71.Memory-augmented Deep Unfolding Network for Guided Image Super-resolution ⬇️

Guided image super-resolution (GISR) aims to obtain a high-resolution (HR) target image by enhancing the spatial resolution of a low-resolution (LR) target image under the guidance of a HR image. However, previous model-based methods mainly takes the entire image as a whole, and assume the prior distribution between the HR target image and the HR guidance image, simply ignoring many non-local common characteristics between them. To alleviate this issue, we firstly propose a maximal a posterior (MAP) estimation model for GISR with two types of prior on the HR target image, i.e., local implicit prior and global implicit prior. The local implicit prior aims to model the complex relationship between the HR target image and the HR guidance image from a local perspective, and the global implicit prior considers the non-local auto-regression property between the two images from a global perspective. Secondly, we design a novel alternating optimization algorithm to solve this model for GISR. The algorithm is in a concise framework that facilitates to be replicated into commonly used deep network structures. Thirdly, to reduce the information loss across iterative stages, the persistent memory mechanism is introduced to augment the information representation by exploiting the Long short-term memory unit (LSTM) in the image and feature spaces. In this way, a deep network with certain interpretation and high representation ability is built. Extensive experimental results validate the superiority of our method on a variety of GISR tasks, including Pan-sharpening, depth image super-resolution, and MR image super-resolution.

72.ModDrop++: A Dynamic Filter Network with Intra-subject Co-training for Multiple Sclerosis Lesion Segmentation with Missing Modalities ⬇️

Multiple Sclerosis (MS) is a chronic neuroinflammatory disease and multi-modality MRIs are routinely used to monitor MS lesions. Many automatic MS lesion segmentation models have been developed and have reached human-level performance. However, most established methods assume the MRI modalities used during training are also available during testing, which is not guaranteed in clinical practice. A training strategy termed Modality Dropout (ModDrop) has been applied to MS lesion segmentation to achieve the state-of-the-art performance for missing modality. We present a novel method dubbed ModDrop++ to train a unified network adaptive to an arbitrary number of input MRI sequences. Moreover, ModDrop++ can be easily applied to any existing model architectures. Specifically, ModDrop++ upgrades the main idea of ModDrop in two key ways. First, we devise a plug-and-play dynamic head and adopt a filter scaling strategy to improve the expressiveness of the network. Second, we design a co-training strategy to leverage the intra-subject relation between full modality and missing modality. In particular, the intra-subject co-training strategy aims to guide the dynamic head to generate similar feature representations between the full- and missing-modality data from the same subject. We use two public MS datasets to show the superiority of ModDrop++. Source code and trained models are available at this https URL.

73.Fluid registration between lung CT and stationary chest tomosynthesis images ⬇️

Registration is widely used in image-guided therapy and image-guided surgery to estimate spatial correspondences between organs of interest between planning and treatment images. However, while high-quality computed tomography (CT) images are often available at planning time, limited angle acquisitions are frequently used during treatment because of radiation concerns or imaging time constraints. This requires algorithms to register CT images based on limited angle acquisitions. We, therefore, formulate a 3D/2D registration approach which infers a 3D deformation based on measured projections and digitally reconstructed radiographs of the CT. Most 3D/2D registration approaches use simple transformation models or require complex mathematical derivations to formulate the underlying optimization problem. Instead, our approach entirely relies on differentiable operations which can be combined with modern computational toolboxes supporting automatic differentiation. This then allows for rapid prototyping, integration with deep neural networks, and to support a variety of transformation models including fluid flow models. We demonstrate our approach for the registration between CT and stationary chest tomosynthesis (sDCT) images and show how it naturally leads to an iterative image reconstruction approach.