Research map / Detection and segmentation / Open-set and few-shot detection
Object-aware Contrastive Learning for Debiased Scene Representation
Sangwoo Mo, Hyunwoo Kang, Kihyuk Sohn, Chun-Liang Li, Jinwoo Shin
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 |
|---|---|---|---|
| add object-aware random crop using ContraCAM boxes | MoCov2 (ResNet-50, COCO-trained) · CIFAR100 linear eval | accuracy (%): 53.26 -> 62.72 | Improved |
| add object-aware random crop using ContraCAM boxes | BYOL (ResNet-50) · CIFAR10 linear eval | accuracy (%): 76.62 -> 82.79 | Improved |
| add OA-Crop with ContraCAM boxes | MoCov2 (ResNet-50) · COCO detection | AP (%): 36.34 -> 36.60 | No clear effect |
| use ground-truth boxes instead of ContraCAM boxes for OA-Crop (ablation) | MoCov2 (ResNet-50) · Food linear eval | accuracy (%): 62.01 -> 57.47 | Hurt |
| use ground-truth boxes instead of ContraCAM boxes for OA-Crop (ablation) | BYOL (ResNet-50) · CIFAR10 linear eval | accuracy (%): 82.79 -> 85.92 | Improved |
| add OA-Crop (CAM) when evaluating on in-distribution ImageNet-9 | MoCov2 (ResNet-50) · ImageNet-9 | accuracy (%): 84.67 -> 84.54 | No clear effect |
| add OA-Crop (CAM) for distribution shift | MoCov2 (ResNet-50) · ImageNet-Sketch-9 | accuracy (%): 41.44 -> 43.11 | Improved |
| add background mixup using ContraCAM | MoCov2 (ResNet-18) · Background Challenge Mixed-Rand | accuracy (%): 60.34 -> 66.89 | Improved |
| add background mixup using ContraCAM | MoCov2 (ResNet-18) · Background Challenge BG-Gap | gap (%): 20.64 -> 17.24 | Improved |
| use BG-Mixup instead of standard Mixup or CutMix (ablation) | MoCov2 (ResNet-18) · ImageNet-C-9 | accuracy (%): 40.15 -> 44.39 | Improved |
| use BG-Mixup instead of Mixup (ablation) | BYOL (ResNet-18) · Stylized-ImageNet-9 | accuracy (%): 28.11 -> 27.01 | Hurt |
| increase ContraCAM iterations from 1 to 10 (ablation) | ContraCAM · COCO mask mIoU | mask mIoU: 0.081 -> 0.319 | Improved |
Closest papers by meaning
- Demystifying Contrastive Self-Supervised Learning: Invariances, Augmentations and Dataset Biases (NeurIPS 2020 · similarity 0.88)
- A Simple Framework for Contrastive Learning of Visual Representations (ICML 2020 · similarity 0.88)
- Revisiting Contrastive Methods for Unsupervised Learning of Visual Representations (NeurIPS 2021 · similarity 0.88)
- When Does Contrastive Visual Representation Learning Work? (CVPR 2022 · similarity 0.87)
- What Should Not Be Contrastive in Contrastive Learning (ICLR 2021 · similarity 0.87)
- Perceptual Inductive Bias Is What You Need Before Contrastive Learning (CVPR 2025 · 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.
