Maserhe / SHAF

Composed Image Retrieval

SHAF for CIR

Setting up

First, clone the repository to a desired location.

Conda Environment

 

 

The following commands will create a local Anaconda environment with the necessary packages installed.

conda create -n shaf -y python=3.8
conda activate shaf
pip install -r requirements.txt

 

Datasets

 

Experiments are conducted on two standard datasets -- Fashion-IQ and SHOES, please see their repositories for download instructions.

 

Training

 

model for training

# Optional: comet experiment logging --api-key and --workspace
python src/combiner_train.py --dataset
dataset_name
--projection-dim
2048
--hidden-dim
4096
--num-epochs
200
--clip-model-name
RN50x4
--combiner-lr
2e-5
--batch-size
512
--clip-bs
32
--transform
targetpad
--target-ratio
1.25
--validation-frequency
1

License

 

License

MIT License applied. In line with licenses from CLIP4Cir and FashionCLIP.

Acknowledgement

Our implementation is based on CLIP4Cir