deformable_detr

DGMN2 with Deformable DETR

This folder contains the implementation of DGMN2 for object detection with Deformable DETR.

Results

COCO validation set

Method	Backbone	Lr schd	AP	Download
Deformable DETR	DGMN2-Tiny	50e	44.4	model
Deformable DETR	DGMN2-Small	50e	47.3	model
Deformable DETR	DGMN2-Medium	50e	48.4	model
Deformable DETR+	DGMN2-Small	50e	48.5	model

Note: "+" indicates using iterative bounding box refinement and two-stage in Deformable DETR.

Getting Started

Clone the repository locally:

git clone https://github.com/fudan-zvg/DGMN2

Installation

a. Install PyTorch and torchvision following the official instructions. Here we use PyTorch 1.8.1 and torchvision 0.9.1.

conda install pytorch==1.8.1 torchvision==0.9.1 cudatoolkit=11.1 -c pytorch -c conda-forge

b. Install PyTorch Image Models. Here we use PyTorch Image Models 0.4.5.

pip install timm==0.4.5

c. Build the extension.

cd dcn
python setup.py build_ext --inplace

d. Install other requirements.

pip install -r requirements.txt

e. Compile CUDA operators in Deformable DETR.

cd ./models/ops
sh ./make.sh
# unit test (should see all checking is True)
python test.py

Data preparation

First, prepare COCO 2017 dataset and organize them as following:

data/
    coco/
        train2017/
        val2017/
        annotations/
            instances_train2017.json
            instances_val2017.json

Then, download the weights pretrained on ImageNet, and put them in a folder pretrained/.

Training

To train DGMN2-Tiny + Deformable DETR on COCO train2017 on a single node with 8 GPUs run:

GPUS_PER_NODE=8 ./tools/run_dist_launch.sh 8 ./configs/dgmn2_deformable_detr.sh --backbone dgmn2_tiny

Evaluation

To evaluate DGMN2-Tiny + Deformable DETR on COCO val2017 on a single node with 8 GPUs run:

GPUS_PER_NODE=8 ./tools/run_dist_launch.sh 8 ./configs/dgmn2_deformable_detr.sh --backbone dgmn2_tiny --resume <path to checkpoint_file> --eval