Evaluation of frozen features

[zhouh02]

Hello, I noticed your paper mentions experiments on the robustness of VGG10, ResNet50, and DINOv2 to changes in viewpoint and lighting. I'd like to ask about some experimental details: What datasets were VGG19 and ResNet pre-trained on? If possible, could you share your evaluation code? Thank you very much! Wishing you success in your research!