This repository contains the training code for the ICML'25 paper titled with "GenZSL: Generative Zero-Shot Learning Via Inductive Variational Autoencoder".
The code implementation of GenZSL mainly based on PyTorch. All of our experiments run and test in Python 3.8.18. To install all required dependencies:
$ pip install -r requirements.txt
You can download the datasets, and organize them as follows:
./dataset
├── data
│ ├── CUB/
│ ├── SUN/
│ └── AWA2/
└── ···
Runing following commands and training GenZSL:
Refer to scripts in ./scripts/usage.sh
Results of our method using various evaluation protocols on three datasets, both in the conventional ZSL (CZSL) and generalized ZSL (GZSL) settings.
| Dataset | Acc(CZSL) | U(GZSL) | S(GZSL) | H(GZSL) |
|---|---|---|---|---|
| CUB | 63.3 | 53.5 | 61.9 | 57.4 |
| SUN | 73.5 | 50.6 | 43.8 | 47.0 |
| AWA2 | 92.2 | 86.1 | 88.7 | 87.4 |
Note: All of above results are run on a server with a NVIDIA TITAN X GPU.