Research map / Detection and segmentation / CNN features
SpeedDETR: Speed-aware Transformers for End-to-end Object Detection
Read the paperCheck 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 |
|---|---|---|---|
| Add scale-combined module (SCM) to HOD+TCM-free model (ablation) | SpeedDETR-tiny · MS COCO val | AP: +1.9 AP (37.8 to 39.7) | Helped |
| Add task-couple module (TCM) (ablation) | SpeedDETR-tiny · MS COCO val | AP: +1.5 AP (39.7 to 41.2) | Helped |
| Replace Swin backbone with hardware-oriented detector (HOD) backbone (ablation) | SpeedDETR-tiny · MS COCO val | AP: +4.3 AP (33.5 to 37.8) at similar GFLOPs (50 to 51) | Helped |
| Apply BN fusion into preceding linear layer (replacing LN-Linear) and RepCNN branch fusion (ablation) | SpeedDETR-tiny · MS COCO val | latency on V100 (ms): 38.1 to 26.2 with both fusions | Helped |
| Use granularity G=2 instead of larger G (ablation) | SpeedDETR-tiny HOD backbone · V100 latency vs AP | AP and speed | Mixed |
| Replace self-attention with convolutional modulation (ConvModulation) | SpeedDETR global attention · ImageNet-1K / V100 speed | inference speed | Helped |
| Replace LN-Linear with Linear-BN in Swin window attention (ablation) | WMSA/SWMSA · ImageNet-1K | speed and accuracy: 13%~15% speedup, <0.3% accuracy degradation | Helped |
Closest papers by meaning
- Speed/Accuracy Trade-Offs for Modern Convolutional Object Detectors (CVPR 2017 · similarity 0.88)
- MobileDets: Searching for Object Detection Architectures for Mobile Accelerators (CVPR 2021 · similarity 0.87)
- SHViT: Single-Head Vision Transformer with Memory Efficient Macro Design (CVPR 2024 · similarity 0.86)
- Can't Slow Me Down: Learning Robust and Hardware-Adaptive Object Detectors against Latency Attacks for Edge Devices (CVPR 2025 · similarity 0.85)
- TGBFormer: Transformer-GraphFormer Blender Network for Video Object Detection (AAAI 2025 · similarity 0.85)
- Context Enhanced Transformer for Single Image Object Detection in Video Data (AAAI 2024 · 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.