Person Attribute Recognition

Refer

• Rethinking of Pedestrian Attribute Recognition: Realistic Datasets and A Strong Baseline
• Bag of Tricks and A Strong Baseline for Deep Person Re-identification
• Omni-Scale Feature Learning for Person Re-Identification
• PyTorch Template Project by victoresque
• Torchreid

Install

• Install anaconda
• conda env create -f environment.yml
• conda activate reid

Dataset

• Automatic download and extract dataset.
• Manual download from github.com/dangweili/pedestrian-attribute-recognition-pytorch
  • PETA

    data_dir/
    |--peta/
    |  |--processed/
    |  |  |--images/
    |  |  |  |--00001.png
    |  |  |  |--00002.png
    |  |  |  |--...
    |  |  |--PETA.mat
    

  • PA-100K

    data_dir/
    |--pa_100k/
    |  |--processed/
    |  |  |--images/
    |  |  |  |--0000001.png
    |  |  |  |--0000002.png
    |  |  |  |--...
    |  |  |--annotation.mat
    

  data_dir in config file.

Run

• python3 train.py --config <path/to/config_file.yml> --colab <true if run on colab else false>

Config

• Using config file in config folder.
• Add new config file based on config/base_epoch.yml or config/base_episode.yml.

Result

Peta dataset

backbone	bn after linear	Head	Loss	mA	Accuracy	Precision	Recall	F1-Score
resnet50	✓	BNHead	CEL_Sigmoid	84.79	80.07	88.28	86.24	86.98
resnet50	✓	BNHead	BCEWithLogitsLoss	79.47	76.33	87.22	82.38	84.33
resnet50_ibn_a_nl	✓	BNHead	CEL_Sigmoid	83.49	79.60	88.89	85.14	86.65
osnet	✓	ReductionHead	CEL_Sigmoid	77.67	73.44	84.17	80.60	81.97
osnet	✓	ReductionHead	BCEWithLogitsLoss	71.00	67.49	85.60	72.94	77.94
osnet	✓	BNHead	CEL_Sigmoid	77.89	72.57	83.68	79.96	81.32
resnet50		BNHead	CEL_Sigmoid	82.67	78.61	88.53	84.17	85.91
resnet50_ibn_a_nl		BNHead	CEL_Sigmoid	82.24	78.57	88.48	84.20	85.91
osnet		ReductionHead	CEL_Sigmoid	77.93	73.00	83.82	80.65	81.81
osnet		BNHead	CEL_Sigmoid	77.72	73.04	84.65	79.82	81.68

PPE dataset

backbone	bn after linear	Head	Loss	mA	Accuracy	Precision	Recall	F1-Score
resnet50	✓	BNHead	CEL_Sigmoid	63.62	85.44	88.28	90.51	88.08

PA-100K

backbone	bn after linear	Head	Loss	mA	Accuracy	Precision	Recall	F1-Score
resnet50	✓	BNHead	CEL_Sigmoid	79.50	78.89	88.17	86.28	86.80

Deploy model with torchserve