Skip to content

Latest commit

 

History

History

pytorch

README.md

Vehicle Re-ID

The code is modified from our baseline code (https://github.com/layumi/Person_reID_baseline_pytorch)

Prerequisite

EfficientNet-Pytorch https://github.com/lukemelas/EfficientNet-PyTorch

Prepare data

Make a dir and put the AICity2020 data into this folder.

mkdir data

Extract XML information https://github.com/PaddlePaddle/Research/tree/master/CV/PaddleReid/process_aicity_data and rename the file.

|- data
    |- 2020AICITY
        |- ...
        |- 000345_c020_9.jpg
        |- ...
        |- 002028_c036_4_9_95_2.jpg

Then you could run the following code to prepare the data for pytorch to load data. You may modify the data path.

python prepare_2020.py            #used to train the re-id model
python prepare_cam2020.py         #used to train the camera-aware model

Train Model

python train_2020.py --name SE_imbalance_s1_384_p0.5_lr2_mt_d0_b24+v+aug   --warm_epoch 5 --droprate 0 --stride 1 --erasing_p 0.5 --autoaug --inputsize 384 --lr 0.02 --use_SE  --gpu_ids 0,1,2  --train_virtual --batchsize 24;

Validate Model

python test_2020.py --name SE_imbalance_s1_384_p0.5_lr2_mt_d0_b24+v+aug

Fine-tune Model

python train_ft_2020.py --name ft_SE_imbalance_s1_384_p0.5_lr2_mt_d0_b24+v+aug  --init_name SE_imbalance_s1_384_p0.5_lr2_mt_d0_b24+v+aug  --droprate 0 --stride 1 --erasing_p 0.5 --inputsize 384 --lr 0.02 --use_SE  --gpu_ids 0,1  --train_all --batchsize 24

Extract Feature for Post-processing

python submit_result_multimodel.py --name ft_SE_imbalance_s1_384_p0.5_lr2_mt_d0_b24+v+aug

If you want to directly test the result, the extracted features & camera prediction & direction prediction could be dowanloaded from GoogleDrive or OneDrive.

Combine the feature extracted from Paddlepaddle & Pytorch Models to output the Submission.

python fast_submit.py