Research map / Detection and segmentation / Open-set and few-shot detection
Region Similarity Representation Learning
Tete Xiao, Colorado J Reed, Xiaolong Wang, Kurt Keutzer, Trevor Darrell
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 region-level similarity learning (ReSim-C4) | MoCo-v2 · VOC detection (IN-100 pretrain) | AP: 54.4 -> 55.9 | Improved |
| add region-level similarity learning (ReSim-C4) | SimSiam · VOC detection (IN-100 pretrain) | AP: 54.6 -> 55.7 | Improved |
| add region-level similarity learning with frozen backbone | MoCo-v2 · VOC detection (IN-100, frozen backbone) | AP: 47.1 -> 50.0 | Improved |
| use a small 16x16 window with stride 16 instead of 48 with stride 32 (ablation) | ReSim-C4 · VOC detection (IN-100 pretrain) | AP: 55.9 -> 55.0 | Hurt |
| use a larger 64 window with stride 48 instead of 48 with stride 32 (ablation) | ReSim-C4 · VOC detection (IN-100, frozen backbone) | AP: 50.0 -> 49.7 | No clear effect |
| use Precise RoI Pooling instead of RoI Align (ablation) | ReSim-C4 · VOC detection (IN-100, frozen backbone) | AP: 48.9 -> 50.0 | Improved |
| use Precise RoI Pooling instead of RoI Align (ablation) | ReSim-C4 · VOC detection (IN-100, full finetune) | AP: 55.7 -> 55.9 | No clear effect |
| add FPN top-down connections to region similarity (ReSim-FPN vs ReSim-C4) (ablation) | ReSim · VOC detection (IN-1K pretrain) | AP: 58.7 -> 59.2 | Improved |
| transfer pretrained FPN weights to the downstream detector | ReSim-FPN · COCO detection 1x | AP bbox: 39.5 -> 39.8 | Improved |
| use FPN-based region similarity instead of C4 (ablation) | ReSim · COCO detection 1x | AP bbox: 39.3 -> 39.5 | Improved |
| extend pretraining from 200 to 400 epochs | ReSim-FPN^T · COCO detection 1x | AP bbox: 39.8 -> 40.3 | Improved |
| replace MoCo-v2 pretraining with ReSim-FPN^T | MoCo-v2 · COCO detection 1x | AP bbox: 38.9 -> 39.8 | Improved |
Closest papers by meaning
- Recognize Any Regions (NeurIPS 2024 · similarity 0.90)
- Region-Based Convolutional Networks for Accurate Object Detection and Segmentation (TPAMI 2016 · similarity 0.89)
- Region-aware Distribution Contrast: A Novel Approach to Multi-Task Partially Supervised Learning (ECCV 2024 · similarity 0.88)
- RegionCL: Exploring Contrastive Region Pairs for Self-Supervised Representation Learning (ECCV 2022 · similarity 0.88)
- REN: Fast and Efficient Region Encodings from Patch-Based Image Encoders (NeurIPS 2025 · similarity 0.87)
- Focus on Local: Finding Reliable Discriminative Regions for Visual Place Recognition (AAAI 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.
