Research map / Detection and segmentation / CNN features
Dense Contrastive Learning for Self-Supervised Visual Pre-Training
Check 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 |
|---|---|---|---|
| add dense local-feature contrastive learning alongside global contrastive learning | MoCo-v2 pre-training · PASCAL VOC object detection | AP: 54.7 -> 56.7 | Helped |
| add dense local-feature contrastive learning alongside global contrastive learning | MoCo-v2 pre-training · COCO object detection | APᵇ: 38.5 -> 39.6 | Helped |
| add dense local-feature contrastive learning alongside global contrastive learning | MoCo-v2 pre-training · COCO instance segmentation | APᵐ: 34.8 -> 35.7 | Helped |
| add dense local-feature contrastive learning alongside global contrastive learning | MoCo-v2 pre-training · PASCAL VOC semantic segmentation | mIoU: 64.5 -> 67.5 | Helped |
| add dense local-feature contrastive learning alongside global contrastive learning | MoCo-v2 pre-training · Cityscapes semantic segmentation | mIoU: 73.8 -> 75.6 | Helped |
| add dense contrastive loss with λ=0.5 (ablation) | MoCo-v2 pre-training · VOC0712 detection | AP: 54.7 -> 56.7 | Helped |
| increase dense-loss weight to λ=0.9 (ablation) | DenseCL · VOC0712 classification | mAP: 82.6 -> 77.8 | Hurt |
| use dense projection features Θ for correspondence matching (ablation) | DenseCL · VOC0712 detection | AP: 56.0 | No clear effect |
| use backbone features F for correspondence matching (ablation) | DenseCL · VOC0712 detection | AP: 56.0 -> 56.7 | Helped |
| increase dense grid size from 1 to 7 (ablation) | DenseCL · VOC0712 detection | AP: 54.6 -> 56.7 | Helped |
| increase dense grid size from 7 to 9 (ablation) | DenseCL · VOC0712 detection | AP: 56.7 -> 56.7 | No clear effect |
| extend pre-training from 200 to 1600 epochs (ablation) | DenseCL · VOC0712 detection | AP: 54.8 -> 57.2 | Helped |
Closest papers by meaning
- Revisiting Contrastive Methods for Unsupervised Learning of Visual Representations (NeurIPS 2021 · similarity 0.89)
- Propagate Yourself: Exploring Pixel-Level Consistency for Unsupervised Visual Representation Learning (CVPR 2021 · similarity 0.89)
- Point-Level Region Contrast for Object Detection Pre-Training (CVPR 2022 · similarity 0.87)
- Self-Contrastive Learning: Single-Viewed Supervised Contrastive Framework Using Sub-network (AAAI 2023 · similarity 0.87)
- Demystifying Contrastive Self-Supervised Learning: Invariances, Augmentations and Dataset Biases (NeurIPS 2020 · similarity 0.87)
- A Simple Framework for Contrastive Learning of Visual Representations (ICML 2020 · similarity 0.87)
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.