Research map / Label-efficient and robust learning / Self-supervised and contrastive
Suppressing Non-Semantic Noise in Masked Image Modeling Representations
Martine Hjelkrem-Tan, Marius Aasan, Rwiddhi Chakraborty, Gabriel Y. Arteaga, Changkyu Choi, Adín Ramírez Rivera
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 SI-score suppression to patch embeddings | CAPI ViT-L14 · ECSSD salient segmentation | max F-beta: 72.456 -> 85.219 | Improved |
| add SI-score suppression to patch embeddings | MAE ViT-B16 · ECSSD salient segmentation | max F-beta: 79.952 -> 82.094 | Improved |
| add SI-score suppression to patch embeddings | DINOv2 ViT-B16 · ECSSD salient segmentation | max F-beta: 71.319 -> 80.633 | Improved |
| add SI-score suppression to patch embeddings | DINOv2 ViT-B16 · DUT-OMRON salient segmentation | max F-beta: 45.923 -> 50.610 | Improved |
| add SI-score suppression to patch embeddings | DINOv2 ViT-B16 · DUT-OMRON salient segmentation | accuracy: 75.067 -> 71.762 | Hurt |
| add SI-score suppression to patch embeddings | DINOv3 ViT-B16 · DUTS salient segmentation | max F-beta: 39.874 -> 47.624 | Improved |
| add SI-score suppression to patch embeddings | MAE ViT-B16 · ImageNet kNN classification | top-1 accuracy: 59.30 -> 60.62 | Improved |
| add SI-score suppression to patch embeddings | DINOv3 ViT-B16 · ImageNet kNN classification | top-1 accuracy: 81.47 -> 81.48 | No clear effect |
| add SI-score suppression to patch embeddings | DINO ViT-B16 (non-MIM) · ADE20k and ImageNet kNN | mIoU: 21.208 -> 21.208 | No clear effect |
| add SI-score suppression to patch embeddings | MAE ViT-B16 · ADE20k kNN segmentation | mIoU: 11.882 -> 12.651 | Improved |
| add Fermi-window scaling of SI-scores (ablation) | SI suppression on CAPI · ImageNet kNN average-pooled | top-1 accuracy: 55.720 -> 56.444 | Improved |
| add Fermi-window scaling of SI-scores (ablation) | SI suppression on I-JEPA · ImageNet kNN average-pooled | top-1 accuracy: 71.422 -> 71.390 | No clear effect |
Closest papers by meaning
- Towards Latent Masked Image Modeling for Self-Supervised Visual Representation Learning (ECCV 2024 · similarity 0.89)
- Beyond [cls]: Exploring the True Potential of Masked Image Modeling Representations (ICCV 2025 · similarity 0.89)
- Pre-training with Random Orthogonal Projection Image Modeling (ICLR 2024 · similarity 0.88)
- On the Role of Discrete Tokenization in Visual Representation Learning (ICLR 2024 · similarity 0.86)
- Wavelet-Driven Masked Image Modeling: A Path to Efficient Visual Representation (AAAI 2025 · similarity 0.86)
- From Prototypes to General Distributions: An Efficient Curriculum for Masked Image Modeling (CVPR 2025 · similarity 0.86)
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.
