Prompt-Based Modality Alignment for Effective Multi-Modal Object Re-Identification

Installation

conda create -n clipreid python=3.8
conda activate clipreid
conda install pytorch==1.8.0 torchvision==0.9.0 torchaudio==0.8.0 cudatoolkit=10.2 -c pytorch
pip install yacs
pip install timm
pip install scikit-image
pip install tqdm
pip install ftfy
pip install regex

Training

For example, if you want to run for the MSVR310, you need to modify the config file to

DATASETS:
  NAMES: ('MSVR310')
  ROOT_DIR: 'your_path'
OUTPUT_DIR: ('output/msvr310')

If you want to run PromptMA:

CUDA_VISIBLE_DEVICES=0 python train.py --config_file configs/msvr310/vit.yml

Evaluation

For example, if you want to test PromptMA for MSVR310

CUDA_VISIBLE_DEVICES=0 python test.py --config_file configs/msvr310/vit.yml TEST.WEIGHT 'your_trained_checkpoints_path/ViT-B-16_120.pth'

Weight

Mould-related weights files can be obtained from this Google Drive

Acknowledgement

Codebase from CLIP-ReID, TransReID, CLIP, and CoOp.