Research map / Detection and segmentation / CNN features
ObjectBox: From Centers to Boxes for Anchor-Free Object Detection
Mohsen Zand, Ali Etemad, Michael Greenspan
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 |
|---|---|---|---|
| augment the center cell with neighboring cells for regression (ablation) | ObjectBox (CSPDarknet) · MS-COCO | AP: 33.1 -> 46.8 | Improved |
| add h-centers to the augmented center locations (ablation) | ObjectBox (CSPDarknet) · MS-COCO | AP: 46.8 -> 41.7 | Hurt |
| regress from 4 box corners instead of center (ablation) | ObjectBox (CSPDarknet) · MS-COCO | AP: 46.8 -> 28.2 | Hurt |
| add the center location to 4 corners (ablation) | ObjectBox (CSPDarknet) · MS-COCO | AP: 28.2 -> 37.4 | Improved |
| predict 4 boxes per location instead of 1 (ablation) | ObjectBox (CSPDarknet) · MS-COCO | AP: 46.8 -> 37.3 | Hurt |
| impose scale constraints m={0,64,128,inf} on feature levels (ablation) | ObjectBox (CSPDarknet) · MS-COCO | AP: 46.8 -> 35.8 | Hurt |
| impose scale constraints m={0,256,512,inf} on feature levels (ablation) | ObjectBox (CSPDarknet) · MS-COCO | AP: 46.8 -> 27.3 | Hurt |
| replace MSE loss with SDIoU loss (ablation) | ObjectBox (CSPDarknet) · MS-COCO | AP: 22.6 -> 46.8 | Improved |
| replace GIoU loss with SDIoU loss (ablation) | ObjectBox (CSPDarknet) · MS-COCO | AP: 27.4 -> 46.8 | Improved |
| replace CIoU loss with SDIoU loss (ablation) | ObjectBox (CSPDarknet) · MS-COCO | AP: 27.1 -> 46.8 | Improved |
| use CSPDarknet backbone instead of ResNet-101 | ObjectBox · MS-COCO | AP: 46.1 -> 46.8 | Improved |
| treat all objects as positives at all scales (ObjectBox) versus FCOS | ObjectBox vs FCOS · MS-COCO | AP: +4 % | Improved |
Later papers that compared with this method
1 later accepted paper reports ObjectBox as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Target-Aware Invertible Encoder with Reconstruction Guidance for Infrared Small Target Detection (CVPR 2026 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FCOS — FCOS: Fully Convolutional One-Stage Object Detection (ICCV 2019) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Unbiased Teacher v2: Semi-Supervised Object Detection for Anchor-Free and Anchor-Based Detectors (CVPR 2022 · similarity 0.90)
- FCOS: A Simple and Strong Anchor-free Object Detector (TPAMI 2020 · similarity 0.89)
- Robust Object Detection with Inaccurate Bounding Boxes (ECCV 2022 · similarity 0.88)
- AABO: Adaptive Anchor Box Optimization for Object Detection via Bayesian Sub-sampling (ECCV 2020 · similarity 0.88)
- Union-over-Intersections: Object Detection beyond Winner-Takes-All (ICLR 2025 · similarity 0.87)
- OpenBox: Annotate Any Bounding Boxes in 3D (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.
