Getting started

Install dependencies

pip install -r requirements.txt

Any components in NLTK can be found here.

Prepare pretrained model

Download pretrained baseline from model_base from BLIP and put model_base.pth at ./checkpoint

Prepare datasets

Download datasets: CUHK-PEDES ICFG-PEDES RSTPReid and put at ./datasets

The files shall be organized as:

|---configs
|---data
|---models
|---datasets
|	---CUHK-PEDES
|	---ICFG-PEDES
|	---RSTPReid
|---checkpoint
|	---model_base.pth
|

Train

bash train.sh

If you have problem in downloading pretrained BERT models here:

tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')

you may manually download it from hugging face and use this instead:

tokenizer = BertTokenizer.from_pretrained(bert_localpath)

Evaluation

bash eval.sh

Acknowledgments

Sincerely appreciate the contributions of BLIP, NLTK, and IRRA, for we used codes and pretrained models from these previous works.