Research map / Detection and segmentation / CNN features
YOLO-MS: Rethinking Multi-Scale Representation Learning for Real-Time Object Detection
Yuming Chen, Xinbin Yuan, Jiabao Wang, Ruiqi Wu, Xiang Li, Qibin Hou, Ming‐Ming Cheng
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 MS-Block instead of Res2Net block (ablation) | RTMDet-T · COCO val ablation (tiny, from scratch) | AP: 40.0 -> 41.0 | Improved |
| add Global Query Learning (GQL) to MS-Block (ablation) | YOLO-MS-XS · COCO val ablation (tiny, from scratch) | AP: 41.0 -> 41.5 | Improved |
| add Heterogeneous Kernel Selection (HKS) protocol to MS-Block + GQL (ablation) | YOLO-MS-XS · COCO val ablation (tiny, from scratch) | AP: 41.5 -> 42.8 | Improved |
| add HKS to MS-Block without GQL (ablation) | YOLO-MS-XS · COCO val ablation (tiny, from scratch) | AP: 41.0 -> 41.3 | Improved |
| increase branches from 2 to 3 to 4 (ablation) | YOLO-MS-XS w/o HKS · COCO val | AP: 39.0 -> 41.5 -> 43.5 | Improved |
| enlarge global query spatial dimension from 2^2 to 4^2 (ablation) | YOLO-MS-XS w/o HKS · COCO val | AP: 41.0 -> 41.5 | Improved |
| enlarge global query beyond 4^2 (5^2, 6^2) (ablation) | YOLO-MS-XS w/o HKS · COCO val | AP: 41.5 -> 41.4 | No clear effect |
| use uniform large kernels instead of 3x3 (ablation) | YOLO-MS-XS · COCO val | AP: 41.0 -> 41.9 | Improved |
| invert HKS order to [9,7,5,3] (large kernels in shallow stages) (ablation) | YOLO-MS-XS · COCO val | AP: 41.9 -> 41.2 | Hurt |
| add HKS in the neck (Neck-HKS) (ablation) | YOLO-MS-XS · COCO val | AP: 41.9 -> 42.3 | Improved |
| apply MS-Block methods to YOLOv8-N | YOLOv8 · COCO val | AP: 37.2 -> 40.2 | Improved |
| apply MS-Block methods to YOLOv10-S | YOLOv10 · COCO val | AP: 46.2 -> 46.8 | Improved |
Closest papers by meaning
- You Only Look Once: Unified, Real-Time Object Detection (CVPR 2016 · similarity 0.96)
- YOLOv10: Real-Time End-to-End Object Detection (NeurIPS 2024 · similarity 0.92)
- YOLO-RD: Introducing Relevant and Compact Explicit Knowledge to YOLO by Retriever-Dictionary (ICLR 2025 · similarity 0.92)
- Mamba YOLO: A Simple Baseline for Object Detection with State Space Model (AAAI 2025 · similarity 0.92)
- YOLO-World: Real-Time Open-Vocabulary Object Detection (CVPR 2024 · similarity 0.90)
- YolOOD: Utilizing Object Detection Concepts for Multi-Label Out-of-Distribution Detection (CVPR 2024 · similarity 0.88)
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.
