Research map / Detection and segmentation / CNN features
LocNet: Improving Localization Accuracy for Object Detection
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 |
|---|---|---|---|
| replace direct box regression with row/column in-out probabilities | Reduced-MR-CNN detection pipeline · VOC2007 test, Edge Box | mAP (IoU ≥ 0.7): 0.570 -> 0.654 | Helped |
| replace direct box regression with row/column in-out probabilities | Reduced-MR-CNN detection pipeline · VOC2007 test, Edge Box | COCO-style mAP: 0.452 -> 0.522 | Helped |
| replace direct box regression with combined border and in-out probabilities | Reduced-MR-CNN detection pipeline · VOC2007 test, Edge Box | mAP (IoU ≥ 0.5): 0.777 -> 0.784 | Helped |
| use border probabilities instead of in-out probabilities (ablation) | LocNet · VOC2007 test, Edge Box, Reduced-MR-CNN | mAP (IoU ≥ 0.7): 0.654 -> 0.644 | Hurt |
| combine border and in-out probabilities rather than use in-out probabilities alone (ablation) | LocNet · VOC2007 test, Edge Box, Reduced-MR-CNN | COCO-style mAP: 0.522 -> 0.530 | Helped |
| replace Edge Box proposals with Selective Search proposals (ablation) | Combined ML with Reduced-MR-CNN · VOC2007 test | COCO-style mAP: 0.530 -> 0.535 | Helped |
| use Combined ML localization instead of no localization | Reduced-MR-CNN with sliding-window candidates · VOC2007 test, sliding windows | mAP (IoU ≥ 0.7): 0.174 -> 0.639 | Helped |
| use Combined ML localization instead of box regression | Reduced-MR-CNN with sliding-window candidates · VOC2007 test, sliding windows | COCO-style mAP: 0.436 -> 0.521 | Helped |
| increase regression region features from 7 x 7 x 512 to 14 x 14 x 512 (ablation) | Bbox reg. · VOC2007 test, Edge Box, Reduced-MR-CNN | mAP (IoU ≥ 0.7): 0.570 -> 0.568 | No clear effect |
| replace the border-probability logistic loss with Gaussian-target mean squared error (ablation) | LocNet Borders ML · VOC detection | localization performance | No clear effect |
| use Combined ML instead of Bbox reg. | Fast R-CNN detection pipeline · COCO 5K mini-val | COCO-style mAP: 0.214 -> 0.264 | Helped |
| use Combined ML instead of Bbox reg. | Fast R-CNN detection pipeline · COCO 5K mini-val | mAP (IoU ≥ 0.75): 0.202 -> 0.282 | Helped |
Closest papers by meaning
- R-FCN: Object Detection via Region-based Fully Convolutional Networks (NeurIPS 2016 · similarity 0.89)
- Multiclass Confidence and Localization Calibration for Object Detection (CVPR 2023 · similarity 0.88)
- HyperNet: Towards Accurate Region Proposal Generation and Joint Object Detection (CVPR 2016 · similarity 0.88)
- Recognize Any Regions (NeurIPS 2024 · similarity 0.87)
- RankDetNet: Delving Into Ranking Constraints for Object Detection (CVPR 2021 · similarity 0.87)
- Learning to Localize Objects with Noisy Labeled Instances (AAAI 2019 · 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.