Research map / Detection and segmentation / CNN features
EfficientDet: Scalable and Efficient Object Detection
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 EfficientNet-B3 (ablation) | RetinaNet with FPN · COCO validation | AP: 37.0 -> 40.3 | Helped |
| Replace FPN with BiFPN (ablation) | EfficientNet-B3 detector · COCO validation | AP: 40.3 -> 44.4 | Helped |
| Add weighted feature fusion (ablation) | BiFPN · COCO validation | AP: 43.94 -> 44.39 | Helped |
| Replace softmax feature fusion with fast normalized fusion (ablation) | Three EfficientDet model sizes · COCO validation | AP: 33.96 -> 33.85 (-0.11); 43.78 -> 43.77 (-0.01); 48.79 -> 48.74 (-0.05) | No clear effect |
| Replace softmax feature fusion with fast normalized fusion (ablation) | Three EfficientDet model sizes · COCO validation | speedup: 1.28x; 1.26x; 1.31x | Helped |
| Jointly scale backbone, BiFPN, prediction network, and input resolution (ablation) | EfficientDet · COCO | efficiency | Helped |
| Train for longer, up to 300 epochs (ablation) | RetinaNet-R50 and EfficientDet-D1 · COCO | AP: +2AP | Helped |
| Use large scale jitter [0.1, 2.0] instead of small jitter [0.8, 1.2] (ablation) | RetinaNet-R50 and EfficientDet-D1 trained for 30 epochs · COCO | AP | Hurt |
| Use large scale jitter [0.1, 2.0] instead of small jitter [0.8, 1.2] (ablation) | RetinaNet-R50 and EfficientDet-D1 trained for 300 epochs · COCO | AP | Helped |
| Scale input resolution together with model dimensions (ablation) | EfficientDet-D1 to D6 · COCO | accuracy-latency trade-off | No clear effect |
| Use EfficientDet-D7x with a larger backbone and an additional feature level | EfficientDet-D7 · COCO test-dev | AP: 53.7 -> 55.1 | Helped |
Closest papers by meaning
- M2Det: A Single-Shot Object Detector Based on Multi-Level Feature Pyramid Network (AAAI 2019 · similarity 0.90)
- Feature Pyramid Networks for Object Detection (CVPR 2017 · similarity 0.88)
- DynamicDet: A Unified Dynamic Architecture for Object Detection (CVPR 2023 · similarity 0.88)
- AugFPN: Improving Multi-Scale Feature Learning for Object Detection (CVPR 2020 · similarity 0.87)
- SpeedDETR: Speed-aware Transformers for End-to-end Object Detection (ICML 2023 · similarity 0.87)
- QueryDet: Cascaded Sparse Query for Accelerating High-Resolution Small Object Detection (CVPR 2022 · similarity 0.86)
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.