AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Hard Patches Mining for Masked Image Modeling

Haochen Wang, Kaiyou Song, Junsong Fan, Yuxi Wang, Jin Xie, Zhaoxiang Zhang

CVPR 2023 · Label-efficient and robust learning › Recognition and long-tail · approach: Correspondence and causal representation learning · 73 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Add loss prediction and learned hard-patch masking (ablation)MAE-style RGB MIM · ImageNet-1KTop-1 accuracy: 82.23 -> 82.95Improved
Add loss prediction objective without using it to generate masks (ablation)MAE-style RGB MIM · ImageNet-1KTop-1 accuracy: 82.23 -> 82.49Improved
Use learned masking with α₀=0 and αT=0.5 rather than random masking (ablation)HPM · ImageNet-1KTop-1 accuracy: 82.49 -> 82.95Improved
Use fully loss-selected masking throughout instead of retaining randomness (ablation)HPM · ImageNet-1KTop-1 accuracy: 82.49 -> 81.40Hurt
Mask patches with the highest predicted loss rather than the lowest (ablation)HPM · ImageNet-1KTop-1 accuracy: 82.36 -> 82.95Improved
Use easy-to-hard rather than hard-to-easy masking (ablation)HPM · ImageNet-1KTop-1 accuracy: 81.71 -> 82.95Improved
Replace absolute MSE loss prediction with relative BCE (ablation)HPM · ImageNet-1KTop-1 accuracy: 82.77 -> 82.95Improved
Use DINO features rather than RGB pixels as the reconstruction target (ablation)HPM · ImageNet-1KTop-1 accuracy: 82.95 -> 84.13Improved
Add HPM to BEiT (ablation)BEiT · ImageNet-1KTop-1 accuracy: 80.9 -> 81.5Improved
Add HPM to iBOT (ablation)iBOT · ImageNet-1KTop-1 accuracy: 82.9 -> 83.4Improved
Add loss prediction and learned masking (ablation)RGB-target pre-training · COCO object detectionAP_box: 40.45 -> 42.03Improved
Add loss prediction and learned masking (ablation)RGB-target pre-training · ADE20k semantic segmentationmIoU: 40.49 -> 42.09Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy