Research map / Label-efficient and robust learning / Self-supervised and contrastive
CASTing Your Model: Learning To Localize Improves Self-Supervised Representations
Ramprasaath R. Selvaraju, Karan Desai, Justin Johnson, Nikhil Naik
Check 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 saliency-constrained crops and Grad-CAM attention loss | MoCo pretrained on COCO · VOC detection | APbbox all: 47.5 -> 54.2 | Improved |
| Add saliency-constrained crops and Grad-CAM attention loss | MoCo pretrained on COCO · VOC classification | mAP: 66.9 -> 73.1 | Improved |
| Add saliency-constrained crops and Grad-CAM attention loss | MoCo pretrained on COCO · ImageNet-1k classification | Top-1 accuracy: 46.5 -> 48.7 | Improved |
| Add saliency-constrained crops and Grad-CAM attention loss | MoCo pretrained on COCO · COCO val | Mean localization IoU: 0.24 -> 0.41 | Improved |
| Add saliency-constrained crops without attention loss (ablation) | MoCo pretrained on COCO · VOC classification | mAP: 66.9 -> 69.3 | Improved |
| Add saliency-constrained crops without attention loss (ablation) | MoCo pretrained on COCO · ImageNet-1k classification | Top-1 accuracy: 46.5 -> 46.0 | Hurt |
| Replace saliency-constrained cropping with random crops (phi=0.0) (ablation) | MoCo-COCO + CAST · VOC classification | mAP: 73.1 -> 72.3 | Hurt |
| Replace saliency-constrained cropping with random crops (phi=0.0) (ablation) | MoCo-COCO + CAST · ImageNet-1k classification | Top-1 accuracy: 48.7 -> 47.4 | Hurt |
| Add attention loss with lambda=3.0 instead of lambda=0.0 (ablation) | MoCo-COCO with constrained crops · VOC classification | mAP: 66.9 -> 73.1 | Improved |
| Increase attention-loss weight to lambda=5.0 from lambda=3.0 (ablation) | MoCo-COCO + CAST · VOC classification | mAP: 73.1 -> 72.4 | Hurt |
| Replace linear projection with a 2-layer MLP (ablation) | MoCo-COCO + CAST · ImageNet-1k classification | Top-1 accuracy: 48.7 -> 50.1 | Improved |
| Restrict attention supervision to the query-key crop intersection (ablation) | MoCo-COCO + CAST · VOC classification | mAP: 73.1 -> 72.0 | Hurt |
Closest papers by meaning
- Attending on Attention ($A^2$): Smaller Self-Supervised ViTs Localize Better Than Larger Ones (NeurIPS 2026 · similarity 0.88)
- Leverage Your Local and Global Representations: A New Self-Supervised Learning Strategy (CVPR 2022 · similarity 0.87)
- Improving Self-Supervised Learning by Characterizing Idealized Representations (NeurIPS 2022 · similarity 0.87)
- Harnessing small projectors and multiple views for efficient vision pretraining (NeurIPS 2024 · similarity 0.86)
- Can Generative Models Improve Self-Supervised Representation Learning? (AAAI 2025 · similarity 0.85)
- On the Surprising Effectiveness of Attention Transfer for Vision Transformers (NeurIPS 2024 · similarity 0.85)
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.
