FHShubho / CGS-ZSL

ChatGPT-guided Semantics for Zero-shot Learning

ChatGPT-guided Semantics for Zero-shot Learning

ChatGPT-guided Semantics for Zero-shot Learning, DICTA 2023.

This paper explores how ChatGPT, a large language model, can enhance class semantics and performance for ZSL tasks. Read: arXiv

Dependencies

Necessary packages can be installed using the environment.yaml file.

To generate the description using the Jupyter Notebook, an API Key is required from OpenAI, which can obtained through registration from here .

Datasets

• Caltech-UCSD Birds-200-2011 (CUB-200-2011)
• Animals with Attributes
• ModelNet
• ScanObjectNN

Model Training

Models are trained following the implementation from the original repository of the papers with slight changes and are listed below.

• DEM
• SynC
• GDAN
• f-CLSWGAN
• TF-VAEGAN
• CADA_VAE

Results

Figure: tSNE visualization of features and semantics of 10 classes from ModelNet40 dataset.

Figure: Results on 2D datasets.

Figure: Results on 3D datasets.

Citation

@inproceedings{CGS-ZSL,
  title={ChatGPT-guided Semantics for Zero-shot Learning},
  author={Fahimul Hoque Shubho, Townim Faisal Chowdhury, Ali Cheraghian, Morteza Saberi, Nabeel Mohammed, Shafin Rahman},
  booktitle={DICTA},
  year={2023}
}