(New!) Official implementation of Parsing R-CNN for Instance-Level Human Analysis (CVPR 2019)
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:
- 8 x TITAN RTX GPU
- pytorch1.1
- python3.6.8
Install Parsing R-CNN following INSTALL.md.
You need to download the datasets and annotations following this repo's formate. As:
-
[MHP-v2](coming soon)
-
DensePoseData(using original MSCOCO2017 images)
And following data structure to train or evaluate Parsing R-CNN models.
On CIHP val
| Backbone | LR | Det AP | mIoU | Parsing (APp50/APvol/PCP50) | DOWNLOAD |
|---|---|---|---|---|---|
| R-50-FPN | 1x | 65.8 | 52.8 | 57.2/51.2/55.4 | |
| R-50-FPN | 3x | 68.7 | 56.0 | 64.1/54.1/60.7 | GoogleDrive |
On MHP-v2 val
| Backbone | LR | Det AP | mIoU | Parsing (APp50/APvol/PCP50) | DOWNLOAD |
|---|---|---|---|---|---|
| R-50-FPN | 1x | 66.5 | 34.0 | 19.9/36.7/32.4 | |
| R-50-FPN | 3x | 69.0 | 36.1 | 27.4/40.5/38.3 | GoogleDrive |
On DensePose_COCO val
| Backbone | LR | Det AP | UV AP (AP/AP50/AP75/APm/APl) | DOWNLOAD |
|---|---|---|---|---|
| R-50-FPN | s1x | 57.4 | 59.3/90.5/68.7/56.2/60.8 | GoogleDrive |
- New metric GPSm is adopted for evaluating UV
ImageNet pretrained weight
coming soon.
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
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
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
Parsing-R-CNN is released under the MIT license.