Research map / Detection and segmentation / CNN features
Gradient Corner Pooling for Keypoint-Based Object Detection
Xuyang Li, Xuemei Xie, Mingxuan Yu, Jiakai Luo, Chengwei Rao, Guangming Shi
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 |
|---|---|---|---|
| Replace corner pooling with gradient corner pooling | Keypoint-based object detectors · MS-COCO | AP: 0.8%, 1.1%, 0.5%, 0.3% | Improved |
| Replace CP with GCP (ablation) | CornerNet-Saccade · MS-COCO val2017 arrayed objects subset (> 15) | AP: 39.5 -> 46.0 | Improved |
| Replace CP with GCP (ablation) | CornerNet-Saccade · MS-COCO val2017 | AP: 42.6 -> 42.8 | Improved |
| Add handcrafted convolution kernels after GCP (ablation) | CornerNet-Saccade · MS-COCO val2017 | AP: 42.8 -> 43.3 | Improved |
| Replace center attention with corner attention (ablation) | CornerNet-Saccade · MS-COCO val2017 | AR100: 62.1 -> 64.2 | Improved |
| Use top-left corner attention instead of center attention (ablation) | CornerNet-Saccade attention maps · MS-COCO val2017 | AP att 30: 52.0 -> 52.8 | Improved |
| Use bottom-right corner attention instead of center attention (ablation) | CornerNet-Saccade attention maps · MS-COCO val2017 | AP att 30: 52.0 -> 53.1 | Improved |
| Use both top-left and bottom-right corner attention maps instead of center attention (ablation) | CornerNet-Saccade attention maps · MS-COCO val2017 | AP att 30: 52.0 -> 54.6 | Improved |
| Increase the maximum number of attention locations from 30 to 40 or 50 (ablation) | CornerNet-Saccade attention mechanism · MS-COCO val2017 | AP att: 52.0; 52.0; 52.0 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CentripetalNet — CentripetalNet: Pursuing High-Quality Keypoint Pairs for Object Detection (CVPR 2020) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- CornerNet: Detecting Objects as Paired Keypoints (ECCV 2018 · similarity 0.90)
- CenterNet: Keypoint Triplets for Object Detection (ICCV 2019 · similarity 0.87)
- Corner Proposal Network for Anchor-free, Two-stage Object Detection (ECCV 2020 · similarity 0.85)
- CentripetalNet: Pursuing High-Quality Keypoint Pairs for Object Detection (CVPR 2020 · similarity 0.85)
- Cornerformer: Purifying Instances for Corner-Based Detectors (ECCV 2022 · similarity 0.84)
- PolarNet: Learning to Optimize Polar Keypoints for Keypoint Based Object Detection (ICLR 2021 · similarity 0.84)
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.
