Research map / Detection and segmentation / Open-set and few-shot detection
Revisiting Contrastive Methods for Unsupervised Learning of Visual Representations
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 |
|---|---|---|---|
| pretrain on scene-centric COCO instead of object-centric ImageNet-118K | MoCo · VOC linear classification | mAP: 75.8 -> 80.9 | Helped |
| pretrain on COCO instead of ImageNet-118K | MoCo · CIFAR10 linear classification | accuracy: 83.1 -> 77.4 | Hurt |
| pretrain on COCO instead of IN-118K, then finetune | MoCo · NYUD semantic seg. | mIoU: 37.7 -> 39.3 | Helped |
| pretrain on long-tailed ImageNet subset instead of uniform | MoCo · VOC linear classification | mAP: 75.8 -> 76.1 | No clear effect |
| add multi-crop (ablation) | MoCo · VOC classification | mAP: 76.1 -> 81.3 | Helped |
| constrain multi-crop to be within the same region (ablation) | MoCo + MC · VOC classification | mAP: 81.3 -> 82.1 | Helped |
| lower momentum (ablation) | MoCo + MC + CC · VOC classification | mAP: 82.1 -> 82.8 | Helped |
| stronger augmentations (ablation) | MoCo + MC + CC + lower m · VOC classification | mAP: 82.8 -> 83.7 | Helped |
| add nearest neighbors positive loss (ablation) | MoCo + MC + CC + lower m + A+ · VOC classification | mAP: 83.7 -> 85.1 | Helped |
| increase neighbors k from 1 to 50 (ablation) | kNN-MoCo · VOC linear | mAP: 83.6 -> 84.3 | Helped |
| raise neighbor loss weight lambda from 0.4 to 0.8 (ablation) | kNN-MoCo · VOC linear | mAP: 84.2 -> 80.7 | Hurt |
| mine neighbors before the projection head (ablation) | kNN-MoCo · VOC linear | mAP: 82.8 -> 84.2 | Helped |
Closest papers by meaning
- Demystifying Contrastive Self-Supervised Learning: Invariances, Augmentations and Dataset Biases (NeurIPS 2020 · similarity 0.91)
- Propagate Yourself: Exploring Pixel-Level Consistency for Unsupervised Visual Representation Learning (CVPR 2021 · similarity 0.91)
- When Does Contrastive Visual Representation Learning Work? (CVPR 2022 · similarity 0.91)
- A Simple Framework for Contrastive Learning of Visual Representations (ICML 2020 · similarity 0.90)
- Dense Contrastive Learning for Self-Supervised Visual Pre-Training (CVPR 2021 · similarity 0.89)
- Momentum Contrast for Unsupervised Visual Representation Learning (CVPR 2020 · similarity 0.89)
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.