Research map / Detection and segmentation / CNN features
Rank & Sort Loss for Object Detection and Instance Segmentation
Kemal Oksuz, Baris Can Cam, Emre Akbas, Sinan Kalkan
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 |
|---|---|---|---|
| replace Focal Loss with RS Loss (no aux head) (ablation) | ATSS · COCO minival | AP: 38.7 -> 39.7 | Improved |
| add score-based weighting to RS Loss (ablation) | ATSS · COCO minival | AP: 39.7 -> 39.8 | No clear effect |
| add value-based task balancing (ablation) | ATSS+RS Loss · COCO minival | AP: 39.8 -> 39.9 | No clear effect |
| remove samplers in RPN and R-CNN (ablation) | RS-R-CNN · COCO minival | AP: 38.5 to 39.6 AP | Improved |
| train Cascade R-CNN with RS Loss | Cascade R-CNN · COCO minival | AP: 40.3 AP to 41.3 AP | Improved |
| replace cross-entropy with RS Loss | Mask R-CNN+RFS · LVIS v1.0 val | mask AP: 21.7 to 25.2 | Improved |
| replace Focal Loss with RS Loss | PAA · COCO minival | AP: 39.9 -> 41.0 | Improved |
| replace Focal Loss with RS Loss | YOLACT · COCO | mask AP: 28.4 -> 29.9 | Improved |
| add OHEM to YOLACT (ablation) | YOLACT · COCO | mask AP: 15.1 -> 28.4 | Improved |
| add size-based normalization to YOLACT (ablation) | YOLACT · COCO | mask AP: 21.7 -> 28.4 | Improved |
| add semantic segmentation head to YOLACT (ablation) | YOLACT · COCO | mask AP: 28.1 -> 28.4 | Improved |
| add sorting error to RS Loss (ablation) | RS-R-CNN · COCO | Spearman rho (IoU vs score): 0.38/0.42 to 0.42/0.47 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- Dynamic R-CNN — Dynamic R-CNN: Towards High Quality Object Detection via Dynamic Training (ECCV 2020) · better on 2 of 2 shared results
- SOLOv2 — SOLOv2: Dynamic and Fast Instance Segmentation (NeurIPS 2020) · better on 1 of 1 shared result
- Libra R-CNN — Libra R-CNN: Towards Balanced Learning for Object Detection (CVPR 2019) · better on 1 of 1 shared result
- Cascade R-CNN — Cascade R-CNN: Delving Into High Quality Object Detection (CVPR 2018) · 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
- Bucketed Ranking-based Losses for Efficient Training of Object Detectors (ECCV 2024 · similarity 0.92)
- DR Loss: Improving Object Detection by Distributional Ranking (CVPR 2020 · similarity 0.90)
- RankDetNet: Delving Into Ranking Constraints for Object Detection (CVPR 2021 · similarity 0.89)
- Revisiting AP Loss for Dense Object Detection: Adaptive Ranking Pair Selection (CVPR 2022 · similarity 0.88)
- AP-Loss for Accurate One-Stage Object Detection (TPAMI 2020 · similarity 0.88)
- Towards Accurate One-Stage Object Detection With AP-Loss (CVPR 2019 · 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.
