Research map / Detection and segmentation / Open-set and few-shot detection
Point-Level Region Contrast for Object Detection 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 |
|---|---|---|---|
| Use point-level region contrast instead of the strongest compared baseline | ResNet-50 pre-training · VOC 07+12 detection, ImageNet-1K pre-training | AP: 59.1 -> 59.4 | Helped |
| Use point-level region contrast instead of the strongest compared baseline | ResNet-50 pre-training · COCO detection, ImageNet-1K pre-training | AP: 40.5 -> 40.7 | Helped |
| Use point-level region contrast instead of the strongest compared baseline | ResNet-50 pre-training · COCO instance segmentation, ImageNet-1K pre-training | AP: 36.4 -> 36.9 | Helped |
| Use point-level region contrast instead of the strongest compared baseline | ResNet-50 pre-training · Cityscapes semantic segmentation, ImageNet-1K pre-training | mIoU: 76.8 -> 77.0 | Helped |
| Use point-level region contrast instead of PixPro | ResNet-50 pre-training · VOC detection, COCO pre-training | AP: 56.5 -> 57.1 | Helped |
| Use point-level region contrast instead of PixPro | ResNet-50 pre-training · COCO detection, COCO pre-training | AP: 39.0 -> 39.8 | Helped |
| Use point-level region contrast instead of PixPro | ResNet-50 pre-training · COCO instance segmentation, COCO pre-training | AP: 35.4 -> 35.9 | Helped |
| Use point-level region contrast instead of PixPro | ResNet-50 pre-training · Cityscapes semantic segmentation, COCO pre-training | mIoU: 75.2 -> 75.9 | Helped |
| Use point-level features rather than pooled region features (ablation) | Point-level region contrast · VOC detection, ImageNet-1K pre-training | AP: 56.5 -> 56.6 | Helped |
| Set the contrastive-versus-affinity loss weight alpha to 0.5 (ablation) | Point-level region contrast · VOC detection, ImageNet-1K pre-training | AP: 57.3 -> 58.0 | Helped |
| Set the image-level-versus-point-level loss weight beta to 0.7 (ablation) | Point-level region contrast · VOC detection, ImageNet-1K pre-training | AP: 56.4 -> 58.0 | Helped |
| Use 32 sampled points per region instead of 16 (ablation) | Point-level region contrast · VOC detection, ImageNet-1K pre-training | AP: 58.0 -> 58.2 | Helped |
Closest papers by meaning
- Aligning Pretraining for Detection via Object-Level Contrastive Learning (NeurIPS 2021 · similarity 0.90)
- Proposal-Contrastive Pretraining for Object Detection from Fewer Data (ICLR 2023 · similarity 0.90)
- Propagate Yourself: Exploring Pixel-Level Consistency for Unsupervised Visual Representation Learning (CVPR 2021 · similarity 0.88)
- Recognize Any Regions (NeurIPS 2024 · similarity 0.88)
- Unsupervised Object Detection Pretraining with Joint Object Priors Generation and Detector Learning (NeurIPS 2022 · similarity 0.88)
- Long-tailed Object Detection Pretraining: Dynamic Rebalancing Contrastive Learning with Dual Reconstruction (NeurIPS 2024 · 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.