Research map / Label-efficient and robust learning / Self-supervised and contrastive
Siamese Image Modeling for Self-Supervised Vision Representation Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use SiameseIM instead of MoCo-v3 | ViT-B/16 SSL pretraining · ImageNet | LIN: 76.7 -> 78.0 | Helped |
| Use SiameseIM instead of MAE | ViT-B/16 SSL pretraining · ImageNet | FT 1%: +14.0 | Helped |
| Use SiameseIM instead of MoCo-v3 | ViT-B/16 SSL pretraining · COCO | APb: +4.2 | Helped |
| Use SiameseIM instead of MAE | ViT-B/16 SSL pretraining · ADE20k | mIoU: +3.0 | Helped |
| Use SiameseIM instead of MAE | ViT-B/16 SSL pretraining · LVIS | APb rare: +1.6 | Helped |
| Use different augmented views rather than the same view (ablation) | SiameseIM · ImageNet | LIN: 62.3 -> 73.1 | Helped |
| Predict features rather than pixels when using different views (ablation) | SiameseIM · ImageNet | LIN: 46.2 -> 69.6 | Helped |
| Add color augmentation with different views (ablation) | SiameseIM · ImageNet | LIN: 69.6 -> 73.1 | Helped |
| Add color augmentation with the same view (ablation) | SiameseIM · ImageNet | LIN: 62.3 -> 59.9 | Hurt |
| Replace LN with BN in projector and decoder (ablation) | SiameseIM · ImageNet | LIN: 73.1 -> 73.6 | Helped |
| Replace random masking with blockwise masking (ablation) | SiameseIM · COCO | APb: 48.7 -> 50.0 | Helped |
| Use MAE-like instead of MoCo-like loss normalization (ablation) | SiameseIM · ImageNet | LIN: 74.7 -> 76.8 | Helped |
Closest papers by meaning
- Siamese Masked Autoencoders (NeurIPS 2023 · similarity 0.90)
- Exploring Simple Siamese Representation Learning (CVPR 2021 · similarity 0.89)
- Harnessing small projectors and multiple views for efficient vision pretraining (NeurIPS 2024 · similarity 0.85)
- Masked Scene Modeling: Narrowing the Gap Between Supervised and Self-Supervised Learning in 3D Scene Understanding (CVPR 2025 · similarity 0.85)
- Contrastive Tuning: A Little Help to Make Masked Autoencoders Forget (AAAI 2024 · similarity 0.85)
- DINOv2 Meets Text: A Unified Framework for Image- and Pixel-Level Vision-Language Alignment (CVPR 2025 · 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.