Awesome

CoT

Code for the paper Hierarchical Visual Primitive Experts for Compositional Zero-Shot Learning, ICCV 2023.

Hanjae Kim, Jiyoung Lee, Seongheon Park, Kwanghoon Sohn

Installation

1. Please create conda environment and install dependencies following the below steps.

conda env create --file environment.yml
conda activate cot

2. Download C-GQA and VAW-CZSL dataset.

• C-GQA
• VAW-CZSL

3. Download Glove word embedding.

• Glove

4. Unzip all downloaded files and place it to the dataset folder following the below structures:

dataset
└─cgqa
│    └─compositional-split-natural
│    └─images
└─vaw-czsl
│    └─...
└─glove
     └─glove.6B.300d.txt
     └─glove_vocab.txt

Training & Testing

1. Update dataset directory in config/*.yml files.
2. To run training code, type

python train.py --cfg config/vaw-czsl.yml

3. For testing, type

python test.py --cfg config/vaw-czsl.yml --load vaw-czsl.pth

Acknowledgement

Our code is based on the following excellent projects;

• ExplanableML/czsl
• nirat1606/OADis
• lukemelas/PyTorch-Pretrained-ViT

Citation

@inproceedings{kim2023hierarchical,
  title={Hierarchical Visual Primitive Experts for Compositional Zero-Shot Learning},
  author={Kim, Hanjae and Lee, Jiyoung and Park, Seongheon and Sohn, Kwanghoon},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={5675--5685},
  year={2023}
}