Research map / Detection and segmentation / Open-set and few-shot detection
Learning to Separate: Detecting Heavily-Occluded Objects in Urban Scenes
Chenhongyi Yang, Vitaly Ablavsky, Kaihong Wang, Qi Feng, Margrit Betke
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 Greedy-NMS with SG-NMS-Linear | Serial R-FCN · KITTI val car (Hard) | AP: 90.21 -> 92.54 | Improved |
| add SG-NMS to the Serial R-FCN baseline (ablation) | Serial R-FCN · KITTI val car (Heavy occlusion) | AP: 44.03 -> 52.62 | Improved |
| add self-attention branch to each head (ablation) | Serial R-FCN · KITTI val car (Heavy occlusion) | AP: 44.03 -> 47.79 | Improved |
| add self-attention branch to each head (ablation) | Serial R-FCN · KITTI val car (Hard) | AP: 89.30 -> 90.10 | Improved |
| add box noise during training (ablation) | Serial R-FCN · KITTI val car (Heavy occlusion) | AP: 44.03 -> 43.98 | No clear effect |
| add box noise during training (ablation) | Serial R-FCN · KITTI val car (Hard) | AP: 89.30 -> 90.01 | Improved |
| add self-attention when SG-NMS is used (ablation) | Serial R-FCN + SG-NMS · KITTI val car (Heavy occlusion) | AP: 52.62 -> 58.43 | Improved |
| add box noise on top of SG-NMS and attention (ablation) | Serial R-FCN + SG-NMS + attention · KITTI val car (Heavy occlusion) | AP: 58.43 -> 61.48 | Improved |
| add SG-NMS on top of noise and attention (ablation) | Serial R-FCN + noise + attention · KITTI val car (Heavy occlusion) | AP: 50.63 -> 61.48 | Improved |
| use semantics-geometry embedding instead of geometric-only embedding (ablation) | SG-NMS · KITTI val car (Heavy occlusion) | AP: 38.12 -> 61.48 | Improved |
| use semantics-geometry embedding instead of semantic-only embedding (ablation) | SG-NMS · KITTI val car (Heavy occlusion) | AP: 55.60 -> 61.48 | Improved |
| use SG-NMS-Square instead of Greedy-NMS best threshold | Serial R-FCN · KITTI val car (Heavy occlusion) | AP: 57.21 -> 62.08 | Improved |
Closest papers by meaning
- Mask-Guided Attention Network for Occluded Pedestrian Detection (ICCV 2019 · similarity 0.89)
- Learning Non-Maximum Suppression (CVPR 2017 · similarity 0.88)
- Precise Detection in Densely Packed Scenes (CVPR 2019 · similarity 0.88)
- WALT: Watch and Learn 2D Amodal Representation From Time-Lapse Imagery (CVPR 2022 · similarity 0.88)
- CorrBEV: Multi-View 3D Object Detection by Correlation Learning with Multi-modal Prototypes (CVPR 2025 · similarity 0.87)
- Roboflow100-VL: A Multi-Domain Object Detection Benchmark for Vision-Language Models (NeurIPS 2025 · similarity 0.86)
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.
