Research map / Detection and segmentation / CNN features
Learning Region Features for Object Detection
Jiayuan Gu, Han Hu, Liwei Wang, Yichen Wei, Jifeng Dai
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 appearance feature term to the geometric term (ablation) | learned region feature module · COCO minival | mAP: 33.2 -> 34.1 | Improved |
| enlarge support region from 1x RoI to whole image (ablation) | learned region feature module · COCO minival | mAP: 33.4 -> 34.3 | Improved |
| enlarge support region from 1x RoI to 2x RoI (ablation) | regular RoI pooling · COCO minival | mAP: 29.8 -> 30.1 | No clear effect |
| enlarge support region from 1x RoI to 2x RoI (ablation) | aligned RoI pooling · COCO minival | mAP: 32.9 -> 32.8 | No clear effect |
| use sparse sampling (196 outside, 14^2 inside) instead of dense enumeration (ablation) | learned region feature module · COCO minival | mAP: 34.2 -> 34.1 | No clear effect |
| sparse sampling with 196 maximum outside positions instead of 100 (ablation) | learned region feature module · COCO minival | mAP: 33.8 -> 34.1 | Improved |
| increase maximum inside sampling positions from 7^2 to 14^2 (ablation) | learned region feature module · COCO minival | mAP: 33.4 -> 34.2 | Improved |
| replace regular RoI pooling by learned region feature module | Faster R-CNN + ResNet-50 · COCO minival | mAP: 29.8 -> 33.4 | Improved |
| replace aligned RoI pooling by learned region feature module | Faster R-CNN + ResNet-50 · COCO minival | mAP: 32.9 -> 33.4 | Improved |
| replace deformable RoI pooling by learned region feature module | Faster R-CNN + ResNet-101 · COCO test-dev | mAP: 36.4 -> 36.4 | No clear effect |
| replace deformable RoI pooling by learned region feature module | FPN + ResNet-101 · COCO test-dev | mAP: 40.0 -> 39.9 | No clear effect |
| replace aligned RoI pooling by learned region feature module | FPN + ResNet-101 · COCO test-dev | mAP: 39.1 -> 39.9 | Improved |
Closest papers by meaning
- Feature Selective Networks for Object Detection (CVPR 2018 · similarity 0.86)
- Deformable Part Region Learning and Feature Aggregation Tree Representation for Object Detection (TPAMI 2023 · similarity 0.85)
- PoolNet+: Exploring the Potential of Pooling for Salient Object Detection (TPAMI 2022 · similarity 0.85)
- Point-Level Region Contrast for Object Detection Pre-Training (CVPR 2022 · similarity 0.85)
- Region-Based Convolutional Networks for Accurate Object Detection and Segmentation (TPAMI 2016 · similarity 0.84)
- Semantic-Aware Transformation-Invariant RoI Align (AAAI 2024 · similarity 0.83)
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.
