Research map / Detection and segmentation / CNN features
FCOS: A Simple and Strong Anchor-free Object Detector
Zhi Tian, Chunhua Shen, Hao Chen, Tong He
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 |
|---|---|---|---|
| add center-ness branch (ablation) | FCOS · COCO val2017 | AP: 38.0 -> 38.9 | Improved |
| compute center-ness from predicted regression vector instead of separate branch (ablation) | FCOS · COCO val2017 | AP: 38.0 -> 37.5 | Hurt |
| use L1 instead of BCE for center-ness loss (ablation) | FCOS · COCO val2017 | AP: 38.9 -> 38.9 | No clear effect |
| add group normalization in heads (ablation) | FCOS · COCO val2017 | AP: 37.9 -> 38.9 | Improved |
| use GIoU loss instead of IoU loss (ablation) | FCOS · COCO val2017 | AP: 38.6 -> 38.9 | Improved |
| use P5 instead of C5 as head input (ablation) | FCOS · COCO val2017 | AP: 38.5 -> 38.9 | Improved |
| center sampling radius 1.5 instead of 1.0 (ablation) | FCOS · COCO val2017 | AP: 38.5 -> 38.9 | Improved |
| center sampling radius 1.5 instead of 2.0 (ablation) | FCOS · COCO val2017 | AP: 38.8 -> 38.9 | No clear effect |
| assign by max(l,t,r,b) instead of original FPN rule (ablation) | FCOS · COCO val2017 | AP: 37.7 -> 38.9 | Improved |
| add FPN multi-level prediction (ablation) | FCOS · COCO val2017 | best possible recall: 96.34 -> 98.95 | Improved |
| add center sampling and FPN (ablation) | FCOS · COCO val2017 | ambiguous sample ratio: 76.60% -> 97.34% | Improved |
| add deformable convolution v2 (ablation) | FCOS ResNeXt-32x8d-101-FPN · COCO test-dev | AP: 44.1 -> 46.6 | Improved |
Closest papers by meaning
- FCOS: Fully Convolutional One-Stage Object Detection (ICCV 2019 · similarity 0.98)
- R-FCN: Object Detection via Region-based Fully Convolutional Networks (NeurIPS 2016 · similarity 0.90)
- Feature Selective Anchor-Free Module for Single-Shot Object Detection (CVPR 2019 · similarity 0.90)
- Learning Rich Features at High-Speed for Single-Shot Object Detection (ICCV 2019 · similarity 0.90)
- Sparse R-CNN: An End-to-End Framework for Object Detection (TPAMI 2023 · similarity 0.89)
- NAS-FCOS: Efficient Search for Object Detection Architectures (IJCV 2021 · similarity 0.89)
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.
