search

OpenGVLab / CaFo

[CVPR 2023] Prompt, Generate, then Cache: Cascade of Foundation Models makes Strong Few-shot Learners
MIT License
343 stars 18 forks source link

cannot reproduce imagenet 16 shots result #7

Closed avaxiao closed 1 year ago

avaxiao commented 1 year ago

Congratulation on your excellent work! It is truly fascinating, and I'm currently conducting experiments based on your code.

However, I'm encountering some difficulties in reproducing your ImageNet 16 shots results. Despite following your config file without making any modifications, the highest test accuracy I've been able to achieve is 68.59, which is slightly lower compared to the reported number of 68.79. I would greatly appreciate any insights or comments you might have regarding this outcome.

ZrrSkywalker commented 1 year ago

Thanks for your interest in our work! +- 0.2% accuracy is within a regular variation range due to machine and environment randomness, such as GPU type, pytorch versions. You may try other random seeds to obtain better results.

avaxiao commented 1 year ago

Thanks!