Research map / Detection and segmentation / Open-set and few-shot detection
Logit Normalization for Long-Tail Object Detection
Liang Zhao, Yao Teng, Limin Wang
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 logit mean subtraction (ablation) | LogN on Mask R-CNN R101 RFS · LVIS v1 val | APb: 27.5 -> 28.7 | Improved |
| add logit std division (ablation) | LogN on Mask R-CNN R101 RFS · LVIS v1 val | APb: 27.5 -> 29.1 | Improved |
| add background calibration scalar beta (ablation) | LogN on Mask R-CNN R101 RFS · LVIS v1 val | APb: 27.5 -> 29.0 | Improved |
| combine mean, std and beta (ablation) | LogN on Mask R-CNN R101 RFS · LVIS v1 val | APb: 27.5 -> 29.8 | Improved |
| set beta to minimum foreground mean instead of background logit mean (ablation) | LogN on Mask R-CNN R101 RFS · LVIS v1 val | APb: 12.1 -> 29.8 | Improved |
| set beta to minimum foreground mean instead of maximum foreground mean (ablation) | LogN on Mask R-CNN R101 RFS · LVIS v1 val | APb: 25.5 -> 29.8 | Improved |
| set beta to minimum instead of average foreground mean (ablation) | LogN on Mask R-CNN R101 RFS · LVIS v1 val | APb: 29.4 -> 29.8 | Improved |
| adaptive beta instead of constant -12 (ablation) | LogN on Mask R-CNN R101 RFS · LVIS v1 val | APb: 29.3 -> 29.8 | Improved |
| online LogN during training with beta=0 instead of post-hoc (ablation) | LogN on LVIS detector · LVIS v1 val | per-group AP | Hurt |
| square the standard deviation in LogN (ablation) | LogN on Mask R-CNN R101 RFS · LVIS v1 val | AP: 23.1% rare AP vs 23.2; total -0.5% | Mixed |
| replace LogA with LogN under RFS | Cascade Mask R-CNN R101 · LVIS v1 val | APb: 32.7 -> 33.8 | Improved |
| replace Seesaw Loss with LogN under RFS | Cascade Mask R-CNN R101 · LVIS v1 val | APb: 32.8 -> 33.8 | Improved |
Closest papers by meaning
- On Model Calibration for Long-Tailed Object Detection and Instance Segmentation (NeurIPS 2021 · similarity 0.87)
- Adaptive Logit Adjustment Loss for Long-Tailed Visual Recognition (AAAI 2022 · similarity 0.87)
- SimLTD: Simple Supervised and Semi-Supervised Long-Tailed Object Detection (CVPR 2025 · similarity 0.87)
- Learning Rich Features at High-Speed for Single-Shot Object Detection (ICCV 2019 · similarity 0.86)
- Long-tailed Object Detection Pretraining: Dynamic Rebalancing Contrastive Learning with Dual Reconstruction (NeurIPS 2024 · similarity 0.85)
- Revisiting Logit Distributions for Reliable Out-of-Distribution Detection (NeurIPS 2025 · similarity 0.84)
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.
