Research map / Detection and segmentation / CNN features
RelationNet++: Bridging Visual Representations for Object Detection via Transformer Decoder
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 |
|---|---|---|---|
| add BVR module (center and corner keys) | RetinaNet R-50 · COCO minival | AP: 35.6 -> 38.5 | Helped |
| add BVR to strong detectors with ResNeXt-64x4d-101-DCN backbone | RetinaNet, Faster R-CNN, FCOS, ATSS · COCO minival | AP: 48.3 -> 50.3 (ATSS) | Helped |
| share key set across feature levels (ablation) | BVR on RetinaNet · COCO minival | AP: 37.0 -> 38.5 | Helped |
| increase number of shared keys from 50 to 200 (ablation) | BVR on RetinaNet · COCO minival | AP: 38.5 -> 38.2 | Hurt |
| use sub-pixel instead of integer center/corner (ablation) | BVR on RetinaNet · COCO minival | AP: 37.0 -> 38.5 | Helped |
| use shared relative location embedding instead of direct computation (ablation) | BVR on RetinaNet · COCO minival | AP: 38.3 -> 38.5 | No clear effect |
| add geometry term to appearance term (ablation) | BVR on RetinaNet · COCO minival | AP: 37.4 -> 38.5 | Helped |
| add appearance term to geometry term (ablation) | BVR on RetinaNet · COCO minival | AP: 37.6 -> 38.5 | Helped |
| use center for classification and corner for regression versus single use (ablation) | BVR on RetinaNet · COCO minival | AP: 37.5 -> 38.5 | Helped |
| use corner keys for classification instead of none (ablation) | BVR on RetinaNet · COCO minival | AP: 35.6 -> 36.2 | Helped |
| add auxiliary point head without using it (multi-task only) (ablation) | RetinaNet R-50 · COCO minival | AP: 35.6 -> 36.4 | Helped |
| add one more head conv layer (deeper RetinaNet) without GN (ablation) | RetinaNet R-50 · COCO minival | AP: 35.6 -> 35.2 | Hurt |
Closest papers by meaning
- R-FCN: Object Detection via Region-based Fully Convolutional Networks (NeurIPS 2016 · similarity 0.90)
- Spatially Invariant Unsupervised Object Detection with Convolutional Neural Networks (AAAI 2019 · similarity 0.89)
- TGBFormer: Transformer-GraphFormer Blender Network for Video Object Detection (AAAI 2025 · similarity 0.89)
- RON: Reverse Connection With Objectness Prior Networks for Object Detection (CVPR 2017 · similarity 0.88)
- Straight to Shapes: Real-Time Detection of Encoded Shapes (CVPR 2017 · similarity 0.88)
- DECO: Unleashing the Potential of ConvNets for Query-based Detection and Segmentation (ICLR 2025 · 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.