PSVMA

Progressive Semantic-Visual Mutual Adaption for Generalized Zero-Shot Learning This repository contains the reference code for the paper "Progressive Semantic-Visual Mutual Adaption for Generalized Zero-Shot Learning" accepted to CVPR 2023.

🌈 Model Architecture

📚 Dependencies

Python 3.6.7
PyTorch = 1.7.0
All experiments are performed with one RTX 3090Ti GPU.

⚡ Prerequisites

Dataset: please download the dataset, i.e., CUB, AWA2, SUN to the dataset root path on your machine
Data split: Datasets can be download from Xian et al. (CVPR2017) and take them into dir ../../datasets/.
Attribute w2v:extract_attribute_w2v_CUB.py extract_attribute_w2v_SUN.py extract_attribute_w2v_AWA2.py should generate and place it in w2v/.
Download pretranined vision Transformer as the vision encoder.

🚀 Train & Eval

Before running commands, you can set the hyperparameters in config on different datasets:

config/cub.yaml       #CUB
config/sun.yaml      #SUN
config/awa2.yaml    #AWA2

T rain:

 python train.py

Eval:

 python test.py

You can test our trained model: CUB, AwA2, SUN.

❗ Cite:

If this work is helpful for you, please cite our paper.

@InProceedings{Liu_2023_CVPR,
    author    = {Liu, Man and Li, Feng and Zhang, Chunjie and Wei, Yunchao and Bai, Huihui and Zhao, Yao},
    title     = {Progressive Semantic-Visual Mutual Adaption for Generalized Zero-Shot Learning},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2023},
    pages     = {15337-15346}
}

📕 Ackowledgement

We thank the following repos providing helpful components in our work. GEM-ZSL

hongbo-sun / PSVMA