MaskPose

DOI : 10.5281/zenodo.14233079

Paper

Rom-Pose is pose estimation model based on simple baseline and ASBU Use restored mask image for increasing accuracy of human pose estimation

Enviroment

Python 3.9.7 CUDA 11.1 relese NVIDIA GPU A5000 used

Download backbone code from URL

https://github.com/microsoft/human-pose-estimation.pytorch - for pose estimation model https://github.com/ducminhkhoi/Amodal-Instance-Seg-ASBU.git - for ASBU model

Installation

1. Follow the Baseline code installation.
2. Change Joints Dataset.py file
3. Change lib/core/function.py
4. Change lib/core/inference.py

Dataset

Download COCO dataset from URL

https://cocodataset.org/#download

${POSE_ROOT}
|-- data
`-- |-- mpii
    `-- |-- annot
        |   |-- gt_valid.mat
        |   |-- test.json
        |   |-- train.json
        |   |-- trainval.json
        |   `-- valid.json
        `-- images
            |-- 000001163.jpg
            |-- 000003072.jpg

Mask image dataset using by data/makedataset.py

How to use

Set all the models need.
Make Whole COCO dataset which consist of answer mask dataset.
Train the ASBU model.
Train the HPE model with ASBU model's output.
Combine together to test.