Research map / Detection and segmentation / CNN features
Deep Reinforcement Learning of Region Proposal Networks for Object Detection
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 |
|---|---|---|---|
| Replace greedy class-agnostic NMS proposal selection with sequential RL-based search and adaptive stopping | Faster R-CNN with VGG-16 · MS COCO 2015 test-dev | mAP@[.5, .95]: 22.3 -> 23.4 | Helped |
| Replace greedy class-agnostic NMS proposal selection with sequential RL-based search and adaptive stopping | Faster R-CNN with VGG-16 · PASCAL VOC 2007 test | mAP@.5: 1.7 mAP above Fr R-CNN | Helped |
| Replace greedy class-agnostic NMS proposal selection with sequential RL-based search and adaptive stopping | Faster R-CNN with VGG-16 · PASCAL VOC 2012 test | mAP@.5: 0.4 mAP | Helped |
| Add posterior class probability adjustment using accumulated class-specific history (ablation) | drl-RPN · MS COCO 2015 test-dev | mAP@[.5, .95]: 0.7 | Helped |
| Add class-specific history and posterior class probability adjustment (ablation) | drl-RPN · PASCAL VOC 2007 test | mAP@.5: 75.2 -> 74.5 | Helped |
| Add class-specific history and posterior class probability adjustment (ablation) | drl-RPN · PASCAL VOC 2012 test | mAP@.5: 70.8 -> 70.4 | Helped |
| Add class-specific history (ablation) | drl-RPN with adaptive stopping · PASCAL VOC 2007 test | mAP@.5: 75.2 -> 74.3 | Helped |
| Add class-specific history (ablation) | drl-RPN with adaptive stopping · PASCAL VOC 2012 test | mAP@.5: 70.8 -> 70.1 | Helped |
| Enforce 12 fixations per image instead of adaptive stopping (ablation) | drl-RPN · PASCAL VOC 2007 test | mAP@.5: 75.2 -> 76.4 | Helped |
| Enforce 12 fixations per image instead of adaptive stopping (ablation) | drl-RPN · PASCAL VOC 2012 test | mAP@.5: 70.8 -> 72.2 | Helped |
| Tune detector parameters jointly with policy learning (ablation) | drl-RPN · PASCAL VOC 2007 test | mAP@.5: 2.0 | Helped |
| Use a preset number of fixations rather than adaptive stopping (ablation) | drl-RPN · PASCAL VOC 2007 test | mAP@.5: 3.0 mAP difference | Mixed |
Closest papers by meaning
- Reinforcement Learning for Visual Object Detection (CVPR 2016 · similarity 0.90)
- Dynamic Zoom-In Network for Fast Object Detection in Large Images (CVPR 2018 · similarity 0.88)
- Deep Variation-Structured Reinforcement Learning for Visual Relationship and Attribute Detection (CVPR 2017 · similarity 0.87)
- R-FCN: Object Detection via Region-based Fully Convolutional Networks (NeurIPS 2016 · similarity 0.87)
- TTRV: Test-Time Reinforcement Learning for Vision Language Models (CVPR 2026 · similarity 0.86)
- RL-SeqISP: Reinforcement Learning-Based Sequential Optimization for Image Signal Processing (AAAI 2024 · similarity 0.85)
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.