Research map / Detection and segmentation / CNN features
MimicDet: Bridging the Gap Between One-Stage and Two-Stage Object Detection
Xin Lu, Quanquan Li, Buyu Li, Junjie Yan
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 T-head without mimic loss (ablation) | MimicDet · COCO minival (ResNet-50, 1x) | AP: 36.0 -> 36.4 | No clear effect |
| add head mimicking loss on top of T-head (ablation) | MimicDet · COCO minival (ResNet-50, 1x) | AP: 36.4 -> 39.0 | Improved |
| replace normal conv alignment layer with guided deformable conv in S-head (ablation) | MimicDet · COCO minival (ResNet-50, 1x) | AP: 33.7 -> 39.2 | Improved |
| replace plain deformable conv with guided deformable conv in S-head (ablation) | MimicDet · COCO minival (ResNet-50, 1x) | AP: 35.2 -> 39.2 | Improved |
| use heavy T-head instead of light T-head without dual-path mimicking (ablation) | MimicDet · COCO minival (S-head AP, ResNet-50) | AP: 37.5 -> 37.1 | Hurt |
| add dual-path (classification and regression) mimicking to heavy T-head (ablation) | MimicDet · COCO minival (S-head AP, ResNet-50) | AP: 37.1 -> 39.0 | Improved |
| use staggered feature pyramid (T-head on P2-P6, S-head on P3-P7) (ablation) | MimicDet · COCO minival (ResNet-50, 1x) | AP: 38.4 -> 39.0 | Improved |
| use shared backbone instead of separated pre-trained teacher backbone (ResNet-50) (ablation) | MimicDet · COCO minival (2x schedule) | AP: 40.2 -> 39.8; training time 34h -> 17.5h | Mixed |
| use shared backbone instead of separated pre-trained teacher backbone (ResNet-101) (ablation) | MimicDet · COCO minival (2x schedule) | AP: 42.2 -> 41.8; training time 47h -> 22h | Mixed |
| increase input size from 400 to 800 (ResNet-50) | MimicDet · COCO minival (2x schedule) | AP: 35.5 -> 39.8 | Improved |
| increase input size from 400 to 800 (ResNet-101) | MimicDet · COCO minival (2x schedule) | AP: 36.1 -> 41.8 | 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.
- FCOS — FCOS: Fully Convolutional One-Stage Object Detection (ICCV 2019) · better on 2 of 2 shared results
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
- Learning Rich Features at High-Speed for Single-Shot Object Detection (ICCV 2019 · similarity 0.88)
- Knowledge Distillation for Object Detection via Rank Mimicking and Prediction-Guided Feature Imitation (AAAI 2022 · similarity 0.88)
- Localization Distillation for Object Detection (TPAMI 2023 · similarity 0.87)
- DetKDS: Knowledge Distillation Search for Object Detectors (ICML 2024 · similarity 0.86)
- MatchDet: A Collaborative Framework for Image Matching and Object Detection (AAAI 2024 · similarity 0.85)
- CrossKD: Cross-Head Knowledge Distillation for Object Detection (CVPR 2024 · similarity 0.85)
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.
