Research map / Detection and segmentation / CNN features
Towards Efficient Use of Multi-Scale Features in Transformer-Based Object Detectors
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 IMFA to DETR-R50 | DETR-R50 · COCO val2017 | AP: 34.9 -> 39.2 | Helped |
| Add IMFA to Conditional-DETR-R50 | Conditional-DETR-R50 · COCO val2017 | AP: 40.9 -> 44.0 | Helped |
| Add IMFA to Anchor-DETR-R50 | Anchor-DETR-R50 · COCO val2017 | AP: 42.1 -> 44.5 | Helped |
| Add IMFA to DAB-DETR-R50 | DAB-DETR-R50 · COCO val2017 | AP: 44.5 -> 45.5 | Helped |
| Add IMFA to DAB-DETR-R50 | DAB-DETR-R50 · COCO val2017 | FLOPs: 202G -> 108G | Mixed |
| Add IMFA to PRTR-R50 at 384x288 rather than using 512x384 input | PRTR-R50 · COCO val2017 human pose estimation | AP kp: 71.0 -> 72.7 | Helped |
| Add IMFA to PRTR-R101 at 384x288 rather than using 512x384 input | PRTR-R101 · COCO val2017 human pose estimation | AP kp: 72.0 -> 73.7 | Helped |
| Use iterative encoder without multi-scale feature sampling (ablation) | DAB-DETR-R50 · COCO val2017 | AP: 42.2 -> 41.9 | Hurt |
| Add sparse multi-scale feature sampling and aggregation to the iterative encoder (ablation) | DAB-DETR-R50 · COCO val2017 | AP: 41.9 -> 45.5 | Helped |
| Replace representative keypoints with random spatial sampling (ablation) | DAB-DETR-R50 + IMFA · COCO val2017 | AP: 45.5 -> 44.2 | Hurt |
| Replace adaptive scale selection with averaged scale selection (ablation) | DAB-DETR-R50 + IMFA · COCO val2017 | AP: 45.5 -> 44.2 | Hurt |
| Remove Dynamic FFN (ablation) | DAB-DETR-R50 + IMFA · COCO val2017 | AP: 45.5 -> 44.7 | Hurt |
Closest papers by meaning
- Feature Aggregated Queries for Transformer-Based Video Object Detectors (CVPR 2023 · similarity 0.86)
- ViDT: An Efficient and Effective Fully Transformer-based Object Detector (ICLR 2022 · similarity 0.85)
- CF-DETR: Coarse-to-Fine Transformers for End-to-End Object Detection (AAAI 2022 · similarity 0.85)
- Lite DETR: An Interleaved Multi-Scale Encoder for Efficient DETR (CVPR 2023 · similarity 0.85)
- TGBFormer: Transformer-GraphFormer Blender Network for Video Object Detection (AAAI 2025 · similarity 0.83)
- Scene Adaptive Sparse Transformer for Event-based Object Detection (CVPR 2024 · similarity 0.83)
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.