instruct-reid

This repo provides a basic training and testing framework for instruction guided person re-identification (instruct-ReID).

Installation

Requirements:
torch==1.8.0+cu111 torchvision==0.9.0+cu111 torchaudio==0.8.0
ftfy==5.8
regex==2023.10.3 
tqdm==4.65.0
transformers==4.31.0
bytecode ==0.15.1
matplotlib==3.8.0
scikit-learn ==1.3.0
opencv-python==4.9.0.80
pyyaml==6.0.1
clip==0.2.0
timm==0.9.16
tensorboardX==2.6.2.2
easydict==1.13
chardet==5.2.0

Prepare Pre-trained Models

shell
mkdir logs && cd logs && mkdir pretrained

download pretrained model pass_vit_base_full.pth and ALBEF.pth to pretrained directory

download pytorch_model.bin to bert-base-uncased

The file tree should be

logs
└── pretrained
    └── pass_vit_base_full.pth
    └── ALBEF.pth
bert-base-uncased
└── pytorch_model.bin

Prepare data

shell
mkdir data

cp the dataset and annotation datalist to data directory.

We provide OmniReID annotation datalist download link

We provide the images original homepage link and please download the CUHK dataset from here, ltcc dataset from here, Market1501 dataset from here, MSMT dataset from here, PRCC dataset from here, COCAS+ Real1 dataset from here, COCAS+ Real2 dataset from here, VC-Clothes dataset from here, LLCM dataset from here, CUHK-PEDES dataset from here and PLIP dataset from here.

The file tree should be

data
└── cuhk
    └── datalist
        └── query.txt
        └── gallery.txt
        └── train.txt
    └── cuhk03_1
└── ltcc
    └── croped_clothes
    └── datalist
        └── query_sc.txt
        └── gallery_sc.txt
        └── query_cc.txt
        └── gallery_cc.txt
        └── query_general.txt
        └── gallery_general.txt
        └── train.txt
    └── LTCC_ReID
    └── templates
    └── white_shirt.jpg
└── market
    └── datalist
        └── query.txt
        └── gallery.txt
        └── train.txt
    └── Market-1501
└── msmt
    └── datalist
        └── query.txt
        └── gallery.txt
        └── train.txt
    └── MSMT17_V1
└── prcc
    └── croped_clothes
    └── datalist
        └── query.txt
        └── gallery.txt
        └── train.txt
    └── rgb
    └── prcc_A_templates
    └── white_shirt.jpg
└── real1
    └── COCAS
    └── datalist
        └── runner_real1_v1_gpt.json
        └── train_attr.txt
        └── train_ctcc.txt
└── real2
    └── real_reid_image_face_blur
    └── datalist
        └── runner_real2_v1_gpt.json
        └── query_attr.txt
        └── gallery_attr.txt
        └── query.txt
        └── gallery.txt
└── vc_clothes
    └── croped_image
    └── datalist
        └── query.txt
        └── gallery.txt
        └── train.txt
    └── gallery
    └── query
    └── train
    └── white_shirt.jpg
└── llcm
    └── LLCM
    └── query.txt
    └── gallery.txt
    └── train.txt
└── CUHK-Pedes
    └── imgs
    └── caption_t2i_v2.json
    └── query_t2i_v2.txt
    └── gallery_t2i_v2.txt
    └── train_t2i_v2.txt
└── PLIP
    └── Part1
    └── Part2
    └── Part3
    └── Part4
    └── Part5
    └── Part6
    └── caption_t2i.json
    └── train_t2i.txt

Training

shell
./scripts/train.sh transformer_dualattn_joint ${gpu_num} ${description} ${port}
# e.g., sh ./scripts/train.sh transformer_dualattn_joint 1 debug 6656

Testing

shell
./scripts/test.sh transformer_dualattn_joint ${/PATH/TO/YOUR/MODEL/} ${test_task_type} ${query-txt} ${gallery-txt} ${root_path} # default 1 GPUs
# e.g., sh ./scripts/test.sh transformer_dualattn_joint <your project root>/Instruct-ReID/checkpoint_cuhk.pth.tar sc <your project root>/Instruct-ReID/data/cuhk/datalist/query.txt <your project root>+/Instruct-ReID/data/cuhk/datalist/gallery.txt <your project root>+/Instruct-ReID/data/cuhk

inference model

We provide inference model for each task at link.

hwz-zju / Instruct-ReID