Research map / Detection and segmentation / CNN features
DetNet: Design Backbone for Object Detection
Zeming Li, Chao Peng, Gang Yu, Xiangyu Zhang, Yangdong Deng, Jian Sun
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 ResNet-50 backbone with DetNet-59 | FPN · COCO minival | mAP: 37.9 -> 40.2 | Improved |
| replace ResNet-101 backbone with DetNet-101 | FPN · COCO minival | mAP: 39.8 -> 41.8 | Improved |
| replace ResNet-50 with DetNet-59 when training from scratch | FPN · COCO minival (from scratch) | mAP: 34.5 -> 36.3 | Improved |
| use DetNet-59 instead of the heavier ResNet-101 (7.6G FLOPs) (ablation) | FPN · COCO minival | mAP: 39.8 -> 40.2 | Improved |
| add 1x1 projection convolution at the start of the new stage (ablation) | DetNet-59 · COCO minival | mAP: 39.1 -> 40.2 | Improved |
| add 1x1 projection convolution at the start of the new stage (ablation) | DetNet-59 · ImageNet val | top-1 error: 24.0 -> 23.5 | Improved |
| pretrain DetNet-59 backbone instead of dilating stage 5 of ResNet-50 at fine-tuning (ablation) | FPN · COCO minival | mAP: 39.0 -> 40.2 | Improved |
| replace ResNet-50 with DetNet-59 for large-object localization | FPN · COCO minival, large objects | AP85: 34.5 -> 40.0 | Improved |
| replace ResNet-50 with DetNet-59 for small-object recall | FPN · COCO minival, small objects | AR50: 60.0 -> 66.4 | Improved |
| replace ResNet-50 with DetNet-59 for small-object localization | FPN · COCO minival, small objects | AR85: 18.7 vs 19.6 | No clear effect |
| replace ResNet-50 with DetNet-59 for large-object recall | FPN · COCO minival, large objects | AR50: 95.0 -> 95.4 | No clear effect |
| replace ResNet-101 with DetNet-59 backbone | FPN · COCO test-dev | mAP: 37.3 -> 40.3 | Improved |
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.
- Faster R-CNN +++ — Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks (TPAMI 2016) · 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
- DetNAS: Backbone Search for Object Detection (NeurIPS 2019 · similarity 0.94)
- CBNet: A Novel Composite Backbone Network Architecture for Object Detection (AAAI 2020 · similarity 0.89)
- Detecting Twenty-Thousand Classes Using Image-Level Supervision (ECCV 2022 · similarity 0.88)
- DetCo: Unsupervised Contrastive Learning for Object Detection (ICCV 2021 · similarity 0.88)
- DetKDS: Knowledge Distillation Search for Object Detectors (ICML 2024 · similarity 0.87)
- MI-DETR: An Object Detection Model with Multi-time Inquiries Mechanism (CVPR 2025 · 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.
