Research map / Label-efficient and robust learning / Self-supervised and contrastive
Bootstrap Masked Visual Modeling via Hard Patch Mining
Haochen Wang, Junsong Fan, Yuxi Wang, Kaiyou Song, Tiancai Wang, Xiangyu Zhang, Zhaoxiang Zhang
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add auxiliary reconstruction loss predictor (ablation) | MAE (RGB) · ImageNet-1K fine-tune | top-1: 82.23 -> 82.49 | Improved |
| add learn-to-mask guided by predicted loss (ablation) | MAE + L_pred · ImageNet-1K fine-tune | top-1: 82.49 -> 82.95 | Improved |
| add HPM to DINO feature distillation (ablation) | DINO features · ImageNet-1K fine-tune | top-1: 83.46 -> 84.13 | Improved |
| add HPM to CLIP feature distillation (ablation) | CLIP features · ImageNet-1K fine-tune | top-1: 83.20 -> 83.58 | Improved |
| add HPM to EMA feature target (ablation) | EMA features · ImageNet-1K fine-tune | top-1: 82.99 -> 83.47 | Improved |
| mask hard patches (argmax) instead of easy patches (argmin) (ablation) | HPM · ImageNet-1K fine-tune | top-1: 82.36 -> 82.95 | Improved |
| use easy-to-hard schedule instead of hard-to-easy (ablation) | HPM · ImageNet-1K fine-tune | top-1: 81.71 -> 82.95 | Improved |
| keep randomness: alpha_T 0.5 instead of 1 (ablation) | HPM · ImageNet-1K fine-tune | top-1: 82.67 -> 82.95 | Improved |
| use relative BCE instead of absolute MSE predictor loss (ablation) | HPM · ImageNet-1K fine-tune | top-1: 82.77 -> 82.95 | Improved |
| use 8 decoder blocks instead of 1 (ablation) | HPM · ImageNet-1K fine-tune | top-1: 82.67 -> 82.95 | Improved |
| add HPM to BEiT | BEiT · ImageNet-1K fine-tune | top-1: 80.9 -> 81.5 | Improved |
| add HPM to MAE (ablation) | MAE (RGB) · COCO detection | AP box: 40.45 -> 42.03 | Improved |
Closest papers by meaning
- Hard Patches Mining for Masked Image Modeling (CVPR 2023 · similarity 0.95)
- SimMIM: A Simple Framework for Masked Image Modeling (CVPR 2022 · similarity 0.86)
- Masked Image Modeling: A Survey (IJCV 2025 · similarity 0.86)
- Towards Latent Masked Image Modeling for Self-Supervised Visual Representation Learning (ECCV 2024 · similarity 0.85)
- From Prototypes to General Distributions: An Efficient Curriculum for Masked Image Modeling (CVPR 2025 · similarity 0.85)
- Morphing Tokens Draw Strong Masked Image Models (ICLR 2025 · similarity 0.84)
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.
