Research map / Detection and segmentation / CNN features
BorderDet: Border Feature for Dense Object Detection
Han Qiu, Yuchen Ma, Zeming Li, Songtao Liu, Jian Sun
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 Border Alignment Module on classification branch (ablation) | FCOS · COCO minival | AP: 38.6 -> 39.7 | Improved |
| add Border Alignment Module on regression branch (ablation) | FCOS · COCO minival | AP90: 16.0 -> 18.6 | Improved |
| add BAM on both branches (ablation) | FCOS · COCO minival | AP: 38.6 -> 41.4 | Improved |
| add BAM on classification branch (ablation) | FCOS · COCO minival | AP90: 16.0 -> 15.9 | No clear effect |
| use border-sensitive feature maps instead of border-agnostic (ablation) | BorderDet · COCO minival | AP: 40.8 to 41.4 | Improved |
| use channel-wise max-pooling instead of channel-wise average-pooling (ablation) | BorderAlign · COCO minival | AP: 40.6 -> 41.4 | Improved |
| use channel-wise instead of border-wise aggregation (max-pooling) (ablation) | BorderAlign · COCO minival | AP: 39.5 -> 41.4 | Improved |
| pooling size 10 instead of 0 (iterative box prediction) (ablation) | BorderAlign · COCO minival | AP: 39.0 -> 41.4 | Improved |
| increase pooling size from 10 to 32 (ablation) | BorderAlign · COCO minival | AP: 41.4 -> 41.4 | No clear effect |
| replace BorderAlign with RoIAlign (ablation) | BorderDet · COCO minival | AP: 41.4 -> 40.4 | Hurt |
| replace BorderAlign with deformable convolution (ablation) | BorderDet · COCO minival | AP: 41.4 -> 39.5 | Hurt |
| add region features to single point feature (ablation) | FCOS · COCO minival | AP: 38.9 -> 39.9 | Improved |
Later papers that compared with this method
2 later accepted papers report BorderDet as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Dynamic DETR: End-to-End Object Detection With Dynamic Attention (ICCV 2021 · better on 2 of 2 shared results)
- CrossDet: Crossline Representation for Object Detection (ICCV 2021 · better on 1 of 1 shared result)
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.
- FCOS — FCOS: Fully Convolutional One-Stage Object Detection (ICCV 2019) · better on 2 of 2 shared results
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
- BORDER: An Oriented Rectangles Approach to Texture-Less Object Recognition (CVPR 2016 · similarity 0.87)
- D2Det: Towards High Quality Object Detection and Instance Segmentation (CVPR 2020 · similarity 0.87)
- Generalized Focal Loss: Towards Efficient Representation Learning for Dense Object Detection (TPAMI 2022 · similarity 0.86)
- AlignDet: Aligning Pre-training and Fine-tuning in Object Detection (ICCV 2023 · similarity 0.85)
- VarifocalNet: An IoU-Aware Dense Object Detector (CVPR 2021 · similarity 0.85)
- CrossDet: Crossline Representation for Object Detection (ICCV 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.
