Research map / Label-efficient and robust learning / Self-supervised and contrastive
Unsupervised Learning of Dense Visual Representations
Pedro O O. Pinheiro, Amjad Almahairi, Ryan Benmalek, Florian Golemo, Aaron C. Courville
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 |
|---|---|---|---|
| Learn pixel-level contrastive representations rather than global image-level representations | VADeR vs. MoCo · PASCAL VOC12 val, frozen features | mIoU: 43.0 -> 56.7 | Improved |
| Learn pixel-level contrastive representations rather than global image-level representations | VADeR vs. MoCo · NYU-depth v2 val, frozen features | RMSE: 1.136 -> 0.964 | Improved |
| Learn pixel-level contrastive representations rather than global image-level representations | VADeR vs. MoCo · COCO val2017 | APbb: 38.9 -> 39.2 | Improved |
| Use correctly corresponding pixels instead of random unmatched pixel pairs (ablation) | VADeR · PASCAL VOC12 val, frozen features | mIoU: 32.4 -> 56.7 | Improved |
| Use correctly corresponding pixels instead of random unmatched pixel pairs (ablation) | VADeR · DAVIS-2017 val, frozen features | J: 9.8 -> 52.4 | Improved |
| Use different crops rather than the same crop for paired views (ablation) | VADeR · PASCAL VOC12 val, frozen features | mIoU: 48.0 -> 56.7 | Improved |
| Use different crops rather than the same crop for paired views (ablation) | VADeR · DAVIS-2017 val, frozen features | J: 46.7 -> 52.4 | Improved |
| Use VADeR initialization instead of supervised ImageNet initialization | Fine-tuning · PASCAL VOC12 val, 2% labels | mIoU: 37.4 -> 47.7 | Improved |
| Use VADeR initialization instead of MoCo initialization | Fine-tuning · PASCAL VOC12 val, 100% labels | mIoU: 75.3 -> 75.4 | No clear effect |
| Use VADeR initialization instead of ImageNet supervised initialization | Fine-tuning · NYU-depth v2 val, 2% labels | RMSE: 1.288 -> 1.128 | Improved |
Closest papers by meaning
- Propagate Yourself: Exploring Pixel-Level Consistency for Unsupervised Visual Representation Learning (CVPR 2021 · similarity 0.88)
- A Simple Framework for Contrastive Learning of Visual Representations (ICML 2020 · similarity 0.88)
- In Pursuit of Pixel Supervision for Visual Pre-training (CVPR 2026 · similarity 0.86)
- Revisiting Self-Supervised Visual Representation Learning (CVPR 2019 · similarity 0.86)
- Towards Efficient and Effective Self-Supervised Learning of Visual Representations (ECCV 2022 · similarity 0.86)
- Dense Siamese Network for Dense Unsupervised Learning (ECCV 2022 · 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.
