Research map / Detection and segmentation / Open-set and few-shot detection
Demystifying Contrastive Self-Supervised Learning: Invariances, Augmentations and Dataset Biases
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 temporally separated frames as positive pairs | MOCOv2 on TrackingNet frames · Pascal VOC classification | mAP: 61.8 -> 63.89 | Helped |
| use temporally separated frames as positive pairs | MOCOv2 on TrackingNet frames · ImageNet linear | top-1: 30.33 -> 29.34 | Hurt |
| add region tracking contrastive loss | MOCOv2 on TrackingNet frames · Pascal Cropped Boxes | mAP: 70.91 -> 75.86 | Helped |
| add region tracking contrastive loss | MOCOv2 on TrackingNet frames · ImageNet linear | top-1: 30.33 -> 36.51 | Helped |
| add region tracking contrastive loss | MOCOv2 on TrackingNet frames · ADE20K | mIoU: 14.69 -> 15.28 | Helped |
| add region tracking contrastive loss | MOCOv2 on TrackingNet frames · Viewpoint invariance | Top-10: 81.55 -> 84.97 | Helped |
| pretrain on object-centric cropped boxes instead of full scene images | MOCOv2 on MSCOCO · Pascal Cropped Boxes | mAP: 71.94 -> 75.29 | Helped |
| pretrain on object-centric cropped boxes instead of full scene images | MOCOv2 on MSCOCO · Pascal VOC classification | mAP: 64.39 -> 59.6 | Hurt |
| contrastive self-supervision vs supervised pretraining | ResNet-50 on ImageNet · Viewpoint invariance | Top-10: 89.54 -> 85.15 | Hurt |
| contrastive self-supervision vs supervised pretraining | ResNet-50 on ImageNet · Occlusion invariance | Top-10: 80.89 -> 84.19 | Helped |
Closest papers by meaning
- Revisiting Contrastive Methods for Unsupervised Learning of Visual Representations (NeurIPS 2021 · similarity 0.91)
- When Does Contrastive Visual Representation Learning Work? (CVPR 2022 · similarity 0.89)
- Self-Supervised Learning of Video-Induced Visual Invariances (CVPR 2020 · similarity 0.89)
- Exploring the Equivalence of Siamese Self-Supervised Learning via a Unified Gradient Framework (CVPR 2022 · similarity 0.89)
- Amortised Invariance Learning for Contrastive Self-Supervision (ICLR 2023 · similarity 0.89)
- A Theoretical Analysis of Self-Supervised Learning for Vision Transformers (ICLR 2025 · 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.