Research map / Detection and segmentation / Open-set and few-shot detection
Benchmarking Object Detectors with COCO: A New Path Forward
Shweta Singh, Aayan Yadav, Jitesh Jain, Humphrey Shi, Justin Johnson, Karan Desai
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 |
|---|---|---|---|
| evaluate on COCO-ReM instead of COCO-2017 | fifty object detectors trained on COCO-2017 · COCO val | mask AP: OneFormer InternImage-H 52.0 -> 59.7 (+7.7) | Improved |
| evaluate on COCO-ReM instead of COCO-2017 for query-based vs region-based rank | Mask2Former Swin-L vs Cascade ViTDet ViT-L · COCO val | mask AP: COCO-2017: 50.0 vs 49.7; COCO-ReM: 55.4 vs 56.4 | Mixed |
| refine mask boundaries with SAM (Stage 1 only) (ablation) | COCO-ReM evaluation set · mean AP of four detectors | mask AP: 49.6 -> 57.3 | Improved |
| add exhaustive instances (LVIS import and LVIS-model masks) (ablation) | COCO-ReM evaluation set · mean AP of four detectors | mask AP: 57.3 -> 55.6 -> 55.5 | No clear effect |
| use higher IoU thresholds for AP (ablation) | COCO-ReM vs COCO-2017 evaluation · AP per IoU threshold | AP | Improved |
| evaluate on COCO-ReM for categories with holes or occlusions | detectors trained on COCO-2017 · per-category AP (80 categories) | AP: higher for 69 of 80 categories, lower for 11 | Mixed |
| train ViTDet-B on COCO-ReM instead of COCO-2017 | ViTDet-B (Mask R-CNN and Cascade) · COCO-ReM val | mask AP | Improved |
| train ViTDet-B on COCO-ReM versus 3x larger ViTDet-L on COCO-2017 | ViTDet-B · COCO-ReM val | mask AP | Improved |
Closest papers by meaning
- COCONut: Modernizing COCO Segmentation (CVPR 2024 · similarity 0.90)
- COCO-O: A Benchmark for Object Detectors under Natural Distribution Shifts (ICCV 2023 · similarity 0.90)
- Object Detection Benchmarks are Incomplete: The Role of Label Errors and Annotation Uncertainty (NeurIPS 2026 · similarity 0.89)
- ContextShift: A Controlled Benchmark for Context Dependence in Object Detection (NeurIPS 2026 · similarity 0.88)
- The Devil is in the Object Boundary: Towards Annotation-free Instance Segmentation using Foundation Models (ICLR 2024 · similarity 0.87)
- Roboflow100-VL: A Multi-Domain Object Detection Benchmark for Vision-Language Models (NeurIPS 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.
