This repository contains the implementation of our approach proposed for Generalized Zero-Shot Learning (GZSL), which combines VAE-GAN-based generative modeling with clustering-based feature selection.
Architecture Diagrams
Ontology schema (img/ontology-schema.png) Semantic model (img/semantic-model.png) Architecture overview (img/architecture-overview.png)
📊 Experimental Results #Zero-Shot Learning (ZSL) #Generalized ZSL (GZSL) #Open-Set GZSL (OS-GZSL)
| Dataset | ZSL | GZSL | OS-GZSL (70-30) | OS-GZSL (50-50) |
|---|---|---|---|---|
| AWA2 | 74.0 | 70.1 | 66.8 | 64.3 |
| CUB | 81.8 | 77.0 | 55.9 | 63.7 |
| SUN | 66.2 | 42.8 | - | - |
| FLO | 91.8 | 92.4 | 84.5 | 79.0 |
📚 Publications
- Akdemir, E., Barisci, N., Akcayol, M.A. et al. Selecting generated synthetic features using clustering algorithm for generalized zero-shot learning. Multimedia Systems 31, 402 (2025). 🔗 https://doi.org/10.1007/s00530-025-01979-z
- Akdemir, E., Barisci, N. Generative-based hybrid model with semantic representations for generalized zero-shot learning. SIViP 19, 27 (2025). 🔗 https://doi.org/10.1007/s11760-024-03734-9
- E. Akdemir and N. Barışçı, “Ontoloji-Based Generalized Zero-Shot Learning with Generative Networks”, GJES, vol. 10, no. 1, pp. 183–192, 2024. 🔗 https://doi.org/10.30855/gmbd.0705n15
Referenced Repositories & Acknowledgements This work builds upon and extends several valuable open-source contributions. We would like to express our sincere thanks to the authors of the following repositories, which we used and/or adapted in the development of our code and experiments:
🔗 https://github.com/akshitac8/tfvaegan – for the base VAE-GAN framework
🔗 https://github.com/uqzhichen/SDGZSL – for semantic description
🔗 https://github.com/genggengcss/OntoZSL – for ontology
🔗 https://github.com/facebookresearch/mixup-cifar10 – for mixup function
In addition, we used the benchmark datasets (AWA2, CUB, FLO) and associated semantic embeddings provided by the above repositories.