Note https://arxiv.org/abs/2303.15166

Note https://refbase.cvc.uab.es/files/MMP2012a.pdf

Note AVA subset for model experiments

Note clip-based; 2022

Note more to evaluate under same accuracy: model bias (shape vs texture); in-distribution vs ood;

facebook/dinov2-with-registers-large

Image Feature Extraction • 0.3B • Updated Dec 23, 2024 • 22k • 9

Note dinov2-reg-large: similar speed vs. convnext-v2-base but slightly lower benchmark qwk; dinov2-reg-base: similar speed vs. convnext-v2-tiny, also slightly higher benchmark qwk; (dinov2 with no register doesn't work well on classification)