Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/josonchan1998/repparser
RepParser: End-to-End Multiple Human Parsing with Representation Parts
https://github.com/josonchan1998/repparser
Last synced: 2 months ago
JSON representation
RepParser: End-to-End Multiple Human Parsing with Representation Parts
- Host: GitHub
- URL: https://github.com/josonchan1998/repparser
- Owner: JosonChan1998
- License: apache-2.0
- Created: 2022-08-12T09:35:04.000Z (over 2 years ago)
- Default Branch: master
- Last Pushed: 2024-06-03T07:36:58.000Z (8 months ago)
- Last Synced: 2024-06-03T09:30:03.228Z (8 months ago)
- Language: Python
- Homepage:
- Size: 901 KB
- Stars: 5
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
- Citation: CITATION.cff
Awesome Lists containing this project
README
# RepParser
ReSParser: Fully Convolutional Multiple Human Parsing With Representative Sets
## Installation
- pytorch 1.7.1
- python 3.7.0
- [mmdetection 2.25.2](https://mmdetection.readthedocs.io/en/latest/get_started.html#installation)
## Dataset
You need to download the datasets and annotations follwing this repo's formate
Make sure to put the files as the following structure:
```
├─data
│ CIHP
│ │ ├─train_img
│ │ ├─train_parsing
│ │ ├─train_seg
│ │ ├─val_img
│ │ ├─val_parsing
│ │ ├─val_seg
│ │ │─annotations
|
├─work_dirs
| ├─resparser_r50_fpn_3x_cihp
| | ├─epoch_75.pth
```
## Results
### CIHP
| Backbone | LR | mIOU | APvol | AP_p50 | PCP50 | download |
|--------------|:----:|:----:|:-----:|:------:|:-----:|:--------:|
| R-50 | 1x | 53.6 | 53.7 | 62.7 | 59.3 |[model](https://drive.google.com/file/d/1IkMpcTjqNtisBZ128AB4kqkTsnklU_04/view?usp=sharing) |
| R-50 | 3x | 57.0 | 55.2 | 66.2 | 62.6 |[model](https://drive.google.com/file/d/1D-R3e_76z_lP23A7W66U16v3C1DGlb9n/view?usp=sharing) |
## Evaluation
```
# inference
CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 ./tools/dist_test.sh configs/ReSParser/resparser_r50_fpn_3x_cihp.py work_dirs/resparser_r50_fpn_3x_cihp/epoch_75.pth 8 --eval bbox --eval-options "jsonfile_prefix=work_dirs/resparser_r50_fpn_3x_cihp/resparser_r50_fpn_3x_cihp_val_result"
# eval, noted that should change the json path produce by previous step.
python utils/eval.py
```
## Training
```
CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 ./tools/dist_train.sh configs/ReSParser/resparser_r50_fpn_3x_cihp.py 8
```
## Citation
```
@misc{chen2022repparser,
title={RepParser: End-to-End Multiple Human Parsing with Representative Parts},
author={Xiaojia Chen and Xuanhan Wang and Lianli Gao and Jingkuan Song},
year={2022},
eprint={2208.12908},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
@ARTICLE{10138431,
author={Dai, Yan and Chen, Xiaojia and Wang, Xuanhan and Pang, Minghui and Gao, Lianli and Shen, Heng Tao},
journal={IEEE Transactions on Multimedia},
title={ReSParser: Fully Convolutional Multiple Human Parsing With Representative Sets},
year={2024},
volume={26},
number={},
pages={1384-1394},
keywords={Pipelines;Task analysis;Kernel;Semantics;Pose estimation;Detectors;Crops;Multiple human parsing;human instance-level analysis;instance-aware modeling},
doi={10.1109/TMM.2023.3281070}}
```