/mot_reid_training

Training Re-Identification model on MOT format

Primary LanguagePython

Training ReID model on MOT format

Setup

pip install -r requirements.txt

Setup environment path

In the [.env] file, rename the variables as following:

  • DATASETS.ROOT_DIR: the path to the dataset directory
  • PRETRAIN_ROOT: the path to the pretrain directory

Data structure:

Please format the reid train from cropped images from MOT dataset . Then put it under the path like the following structure:

data/
├── person_reid/
│   ├── gallery/
│   ├── query/
│   └── train/
└──

Pretrained on ImageNet

Put the pretrained model like the following structure

├── datasets/
│   ├── reid/
│   ├── pretrain/
│       ├── resnet50-19c8e357.pth
│       ├── -------
│       └── jx_vit_base_p16_224-80ecf9dd.pth

Training ReID

bash run.sh

Before training create logs/ folder to save status from every running state .After training, the weight will be stored in the lightning_logs/ folder. Navigate to this folder and copy the corresponding epoch weight of each model to the corresponding folder in output/weight.