Research map / Detection and segmentation / Open-set and few-shot detection
"A Simple Approach and Benchmark for 21,000-Category Object Detection"
Yutong Lin, Chen Li, Yue Cao, Zheng Zhang, Jianfeng Wang, Lijuan Wang, Zicheng Liu, Han Hu
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 reformulated classification (RoI head, RoIAlign) (ablation) | two-stage approach, ResNet-50 · LVIS-997-novel | AP mask novel: 21.8 -> 22.8 | Improved |
| add stage I classification training (ablation) | two-stage approach, ResNet-50 · LVIS-997-novel | AP mask novel: 13.0 -> 22.8 | Improved |
| add reformulated classification (ablation) | two-stage approach, ResNet-50 · ImageNet-1K | Top-1 localization accuracy: 60.8 -> 64.7 | Improved |
| add stage I classification training (ablation) | two-stage approach, ResNet-50 · ImageNet-1K | Top-1 localization accuracy: 50.4 -> 64.7 | Improved |
| add reformulated classification (ablation) | two-stage approach, ResNet-50 · ImageNet-1K | GT-Known accuracy: 69.9 -> 73.8 | Improved |
| add stage I classification training (ablation) | two-stage approach, ResNet-50 · ImageNet-1K | GT-Known accuracy: 71.0 -> 73.8 | Improved |
| scale backbone from ResNet-50 to Swin-S (ablation) | two-stage approach · ImageNet-1K | Top-1 localization accuracy: 64.7 -> 68.3 | Improved |
| scale backbone from ResNet-50 to Swin-S | two-stage approach · LVIS-997 | AP mask: 27.6 -> 35.1 | Improved |
| scale backbone from Swin-T to Swin-S | two-stage approach · LVIS-997 | Delta AP (novel, all): -1.6 -> -4.5 | Hurt |
| use reverse flow (classifier first) instead of Detic | detector · ImageNet-21K benchmark | AP50: 1.3 -> 6.9 | Improved |
| successive IN-21K -> O365 training instead of classification-only pre-training | LVIS fine-tuning · LVIS | AP: 35.8 -> 38.8 | Improved |
| joint training instead of successive training as pre-training (ablation) | LVIS fine-tuning · LVIS | AP: 38.8 -> 40.1 | Improved |
Closest papers by meaning
- Described Object Detection: Liberating Object Detection with Flexible Expressions (NeurIPS 2023 · similarity 0.91)
- Simple Open-Vocabulary Object Detection with Vision Transformers (ECCV 2022 · similarity 0.91)
- Towards Large-Scale Small Object Detection: Survey and Benchmarks (TPAMI 2023 · similarity 0.91)
- Object Detection Benchmarks are Incomplete: The Role of Label Errors and Annotation Uncertainty (NeurIPS 2026 · similarity 0.89)
- SimLTD: Simple Supervised and Semi-Supervised Long-Tailed Object Detection (CVPR 2025 · similarity 0.89)
- How to Evaluate the Generalization of Detection? A Benchmark for Comprehensive Open-Vocabulary Detection (AAAI 2024 · 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.
