Reimplementation_SSC

Pytorch unofficial reimplementation for "Spatial and semantic consistency regularizations for pedestrian attribute recognition", ICCV2021

Results

PETA	mA	Acc	Prec	Recall	F1
Paper SSC	86.52	78.95	86.02	87.12	86.99
Unofficial SSC	83.90	78.60	86.16	86.59	86.38

PA100K	mA	Acc	Prec	Recall	F1
Paper SSC	81.87	78.89	85.98	89.10	86.87
Unofficial SSC	79.51	78.49	86.80	87.30	87.05

RAP	mA	Acc	Prec	Recall	F1
Paper SSC	82.77	68.37	75.05	87.49	80.43
Unofficial SSC	82.14	68.16	77.87	82.88	79.87

Data

You can download dataset from "https://github.com/valencebond/Rethinking_of_PAR".

${POSE_ROOT}
 |-- data
     |-- peta
         |-- images/
         |-- dataset_all.pkl
     |-- rap
         |-- RAP_dataset/
         |-- dataset_all.pkl
     |-- pa100k
         |-- data/
         |-- dataset_all.pkl

Training

python tools/train.py --cfg experiments/[dataset name].yaml --gpu 0 -- savename [save path of outputs]

Acknowledgements

Thanks for their open-source codes HRNet and Jian Jia.