ReID_baseline

A strong baseline (state-of-the-art) for person re-identification.

We support

• easy dataset preparation
• end-to-end training and evaluation
• multi-GPU distributed training
• fast data loader with prefetcher
• fast training speed with fp16
• fast evaluation with cython
• support both image and video reid
• multi-dataset training
• cross-dataset evaluation
• high modular management
• state-of-the-art performance with simple model
• high efficient backbone
• advanced training techniques
• various loss functions
• tensorboard visualization

Get Started

The designed architecture follows this guide PyTorch-Project-Template, you can check each folder's purpose by yourself.

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

2. Run git clone https://github.com/L1aoXingyu/reid_baseline.git

3. Install dependencies:

   • pytorch 1.0.0+
   • torchvision
   • tensorboard
   • yacs

4. Prepare dataset

   Create a directory to store reid datasets under this repo via

   cd reid_baseline
   mkdir datasets

   1. Download dataset to datasets/ from baidu pan or google driver
   2. Extract dataset. The dataset structure would like:

   datasets
       Market-1501-v15.09.15
           bounding_box_test/
           bounding_box_train/

5. Prepare pretrained model. If you use origin ResNet, you do not need to do anything. But if you want to use ResNet_ibn, you need to download pretrain model in here. And then you can put it in ~/.cache/torch/checkpoints or anywhere you like.

   Then you should set this pretrain model path in configs/softmax_triplet.yml.

6. compile with cython to accelerate evalution

   cd csrc/eval_cylib; make

Train

Most of the configuration files that we provide, you can run this command for training market1501

bash scripts/train_openset.sh

Or you can just run code below to modify your cfg parameters

CUDA_VISIBLE_DEVICES='0,1' python tools/train.py -cfg='configs/softmax_triplet.yml' DATASETS.NAMES '("dukemtmc","market1501",)' SOLVER.IMS_PER_BATCH '256'

Test

You can test your model's performance directly by running this command

CUDA_VISIBLE_DEVICES='0' python tools/test.py -cfg='configs/softmax_triplet.yml' DATASET.TEST_NAMES 'dukemtmc' \
MODEL.BACKBONE 'resnet50' \
MODEL.WITH_IBN 'True' \
TEST.WEIGHT '/save/trained_model/path'

Experiment Results

size=(256, 128) batch_size=64 (16 id x 4 imgs)
softmax?	✔︎	✔︎	✔︎	✔︎
label smooth?			✔︎	✔︎
triplet?		✔︎	✔︎	✔︎
ibn?			✔︎	✔︎
gcnet?				✔︎
Market1501	93.4 (82.9)	94.2 (86.1)	95.4 (87.9)	95.2 (88.7)
DukeMTMC-reid	84.7 (72.7)	87.3 (76.0)	89.5 (79.7)	90.0 (80.2)
CUHK03

🔥Any other tricks are welcomed！