Research map / Detection and segmentation / Open-set and few-shot detection
Universal Object Detection with Large Vision Model
Feng Lin, Wenze Hu, Yaowei Wang, Yonghong Tian, Guangming Lu, Fanglin Chen, Yong Xu, Xiaoyu Wang
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 |
|---|---|---|---|
| use Cascade R-CNN heads instead of Faster R-CNN (ablation) | frozen SEER-RegNet32gf detector · COCO val | mAP: 29.0 -> 39.9 | Improved |
| add stacked NAS-FPN neck to Cascade R-CNN (ablation) | Cascade R-CNN · COCO val | mAP: 39.9 -> 45.7 | Improved |
| add Cascade RPN on top of stacked NAS-FPN (ablation) | Cascade R-CNN + NAS-FPN · COCO val | mAP: 45.7 -> 47.6 | Improved |
| stack more BiFPN blocks (ablation) | Cascade R-CNN + BiFPN · COCO val | mAP: 41.5 -> 41.1 | Hurt |
| freeze the SEER backbone instead of finetuning (ablation) | Cascade R-CNN + NAS-FPN + Cascade RPN · COCO val | mAP: 49.9 -> 51.4 | Improved |
| freeze the SEER-RegNet256gf backbone to cut memory (ablation) | detector with SEER-RegNet256gf · MVD | GPU memory (GB/image): 60 -> 15 | Improved |
| add HCLS to OID hierarchy loss (ablation) | Cascade R-CNN (32gf) · COCO val | mAP: 36.4 -> 37.1 | Improved |
| add HCLS to OID hierarchy loss (ablation) | Cascade R-CNN (32gf) · OID val | AP50: 64.8 -> 65.2 | Improved |
| add HCLS to OID hierarchy loss (ablation) | Cascade R-CNN + NAS-FPN + Cascade RPN · OID val | AP50: 67.6 -> 68.0 | Improved |
| use unified hierarchy loss instead of baseline (ablation) | Cascade R-CNN + NAS-FPN + Cascade RPN · COCO val | mAP: 44.1 -> 43.1 | Hurt |
| use naive loss suppression instead of baseline (ablation) | Cascade R-CNN (32gf) · OID val | AP50: 61.7 -> 62.7 | Improved |
| train universal detector on all three datasets instead of individually (ablation) | Large-UniDet [S] · COCO val | mAP: 47.6 -> 48.8 | Improved |
Closest papers by meaning
- Detecting Everything in the Open World: Towards Universal Object Detection (CVPR 2023 · similarity 0.87)
- Towards Universal Object Detection by Domain Attention (CVPR 2019 · similarity 0.86)
- Perspective-Invariant 3D Object Detection (ICCV 2025 · similarity 0.85)
- UniDetector: Towards Universal Object Detection With Heterogeneous Supervision (TPAMI 2024 · similarity 0.85)
- Cascade-DETR: Delving into High-Quality Universal Object Detection (ICCV 2023 · similarity 0.85)
- Efficient Test-time Adaptive Object Detection via Sensitivity-Guided Pruning (CVPR 2025 · similarity 0.83)
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.
