wwn1233 / beyond-strong-baseline-for-person-reid

Beyond strong basleine for Person Re-ID. 2019

Beyond strong basleine for Person Re-ID. 2019

The codes follow the structure of Detectron.pytorch and AlignedReID. It builds a very strong baseline for Re-Identification task by exploring bag of tricks. Without center loss and label smoothing, our project can achieve comparable(or better) results than reid-strong-baseline. Also, the project serves as the baseline of our new method，denoted as Orthogonal Center Learning with Subspace Masking. We are grateful for your contribution on this project and hope the project can help your research or work.

Authors

Weinong Wang

Supported tricks

• Warm up learning rate (wu)
• Last stride = 1 (lsr)
• BNNeck (bnn)
• BN of zero gamma (bnzg)
• classifier have no bias (cnb)
• Random erasing augmentation (re)
• Label smoothing (lsm)
• Center loss (cl)
• Re-ranking (rr)
• backbone: ResNet-50, ResNet-18

Supported dataset

• Market1501
• DukeMTMC-reID
• CUHK03
• MSMT17

Experements results (rank1/mAP)

Note: Although the tricks mentioned above are all supported, our baseline does not have all of them.

Model	Market1501	DukeMTMC-reID	CUHK03	MSMT17
reid-strong-baseline (ResNet-50+wu+lsr+bnn+re+lsm+cl)	94.5 (85.9)	86.4 (76.4)	-	-
ours(ResNet-50+wu+lsr+bnn+re)	93.9 (84.6)	86.6 (77.1)	73.1 (70.6)	80.6(65.1)
ours(ResNet-50+wu+lsr+bnn+re+cnb)	93.8 (84.8)	87.1 (77.3)	76.2 (52.9)	75.9(52.6)

Requirements

Training and Testing on Python3.5

pytorch = 0.4.0
torchvision>=0.2.0
matplotlib
numpy
scipy
opencv
pyyaml
packaging
PIL
tqdm
time

Get Started

1. cd to folder where you want to download this repo

2. Run git clone https://github.com/paozhuanyinyuba/Beyond-Strong-Baseline-for-Person-ReID.git

3. Prepare dataset

   Create a directory to store reid datasets under this repo or outside this repo. The path to the root of the dataset is set by the command line argument, namely “ - - root **” .

   For instance, you can create a directory to store reid datasets under this repo via

   cd Beyond-Strong-Baseline-for-Person-ReID
   mkdir data

   （1）Market1501

   • Download dataset to data/ from http://www.liangzheng.org/Project/project_reid.html
   • Extract dataset and rename to market1501. The data structure would like:

   data
       market1501 # this folder contains 6 files.
           bounding_box_test/
           bounding_box_train/
           ......

   （2）DukeMTMC-reID

   • Download dataset to data/ from https://github.com/layumi/DukeMTMC-reID_evaluation#download-dataset
   • Extract dataset and rename to dukemtmc-reid. The data structure would like:

   data
       dukemtmc-reid
       	DukeMTMC-reID # this folder contains 8 files.
           	bounding_box_test/
           	bounding_box_train/
           	......

   （3）CUHK03

   • Download dataset to data/ from http://www.ee.cuhk.edu.hk/~xgwang/CUHK_identification.html
   • Extract dataset and rename to cuhk03. The data structure would like:

   data
       cuhk03
       	cuhk03_new_protocol_config_labeled.mat #
       	cuhk03_new_protocol_config_detected.mat #
       	cuhk03_release #
           	cuhk-03.mat
           	README.md

   （4）MSMT17

   • Download dataset to data/ from https://www.pkuvmc.com/publications/msmt17.html
   • Extract dataset and rename to msmt17. The data structure would like:

   data
       msmt17
       	MSMT17_V1 # this folder contains 6 files.
           	train/
           	test/
           	......

4. Prepare pretrained model

   Create a directory to store pretrained model under this repo or outside this repo. The path to the pretrained model is set in every single training config file in configs/reid/*.yaml. Take configs/reid/R-50-Base.yaml for example, line 6 shows: "RESNETS: IMAGENET_PRETRAINED_WEIGHTS: 'pretrained_model/resnet50-19c8e357.pth'"

   For instance, you can create a directory to store the pretrained model under this repo via

   cd Beyond-Strong-Baseline-for-Person-ReID
   mkdir pretrained_model

   （1）ResNet

   • put ResNet-50 or ResNet-18 in the folder pretrained_model/

   （2）More backbone coming soon ......

5. If you want to know the detailed configurations and their meaning, please refer to lib/core/config.py.

Train

• ResNet-50+wu+lsr+bnn+re

python3 reid_tools/train_net_step_reid.py --dataset "('dataset name, eg. cuhk03, market1501, msmt17 and dukemtmcreid')"  --root "('Your path to the root of the dataset')" --eval-step  "('How many epochs to test once')" --cfg configs/reid/R-50-Base.yaml  --save-dir "('your path to save checkpoints and logs')"

• ResNet-18+wu+lsr+bnn+re

python3 reid_tools/train_net_step_reid.py --dataset "('dataset name, eg. cuhk03, market1501, msmt17 and dukemtmcreid')"  --root "('Your path to the root of the dataset')" --eval-step  "('How many epochs to test once')" --cfg configs/reid/R-18-Base.yaml  --save-dir "('your path to save checkpoints and logs')"

Test

• ResNet-50+wu+lsr+bnn+re. Test with Euclidean distance using feature before BN without re-ranking.

python3 reid_tools/train_net_step_reid.py --save-dir "('your path to save checkpoints and logs')"  --cfg configs/reid/R-50-Base.yaml --root "('Your path to the root of the dataset')"  --dataset "('dataset name, eg. cuhk03, market1501, msmt17 and dukemtmcreid')"  --evaluate --load_ckpt "('your path to trained checkpoints')"

• ResNet-18+wu+lsr+bnn+re. Test with Euclidean distance using feature before BN without re-ranking.

python3 reid_tools/train_net_step_reid.py --save-dir "('your path to save checkpoints and logs')"  --cfg configs/reid/R-18-Base.yaml --root "('Your path to the root of the dataset')"  --dataset "('dataset name, eg. cuhk03, market1501, msmt17 and dukemtmcreid')"  --evaluate --load_ckpt "('your path to trained checkpoints')"