bramtoula/vtab_caltech101

VTAB Caltech101

This dataset has been used for the paper Fantastic Features and Where to Find Them: A Probing Method to combine Features from Multiple Foundation Models (NeurIPS 2025).

It reproduces the settings (splits, labels) used for the Visual Task Adaptation Benchmark (VTAB).

• VTAB Paper: A Large-scale Study of Representation Learning with the Visual Task Adaptation Benchmark
• VTAB Repository: google-research/task_adaptation

Details of the original dataset:

• Original Citation: Fei-Fei, Li, Robert Fergus, and Pietro Perona. "One-shot learning of object categories." IEEE Transactions on Pattern Analysis and Machine Intelligence.
• Caltech 101 Homepage: Caltech101 on CaltechDATA