Parsing-R-CNN

(New!) Official implementation of Parsing R-CNN for Instance-Level Human Analysis (CVPR 2019)

Citing Parsing R-CNN

If you use Parsing R-CNN, please use the following BibTeX entry.

@inproceedings{yang2019cvpr,
  title = {Parsing R-CNN for Instance-Level Human Analysis},
  author = {Lu Yang and Qing Song and Zhihui Wang and Ming Jiang},
  booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year = {2019}
}

In this repository, we release the Parsing R-CNN code in Pytorch.

• Parsing R-CNN architecture:

• Parsing R-CNN output:

Installation

• 8 x TITAN RTX GPU
• pytorch1.1
• python3.6.8

Install Parsing R-CNN following INSTALL.md.

Dataset

You need to download the datasets and annotations following this repo's formate. As:

• CIHP

• [MHP-v2](coming soon)

• DensePoseData(using original MSCOCO2017 images)

And following data structure to train or evaluate Parsing R-CNN models.

Results and Models

On CIHP val

On MHP-v2 val

On DensePose_COCO val

• New metric GPSm is adopted for evaluating UV

ImageNet pretrained weight

• R-50
• R-50-GN
• X-101-32x8d

Visualization

coming soon.

Training

To train a model with 8 GPUs run:

python -m torch.distributed.launch --nproc_per_node=8 tools/train_net.py --cfg cfgs/CIHP/e2e_rp_rcnn_R-50-FPN_3x_ms.yaml

Evaluation

multi-gpu evaluation,

python tools/test_net.py --cfg ckpts/CIHP/e2e_rp_rcnn_R-50-FPN_3x_ms/e2e_rp_rcnn_R-50-FPN_3x_ms.yaml --gpu_id 0,1,2,3,4,5,6,7

single-gpu evaluation,

python tools/test_net.py --cfg ckpts/CIHP/e2e_rp_rcnn_R-50-FPN_3x_ms/e2e_rp_rcnn_R-50-FPN_3x_ms.yaml --gpu_id 0

License

Parsing-R-CNN is released under the MIT license.