Research map / Detection and segmentation / CNN features
Lite DETR: An Interleaved Multi-Scale Encoder for Efficient DETR
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 |
|---|---|---|---|
| Replace DINO's encoder with the Lite interleaved encoder and KDA | DINO with Swin-T backbone · COCO val2017 | AP: 54.1 -> 53.9 | Hurt |
| Replace DINO's encoder with the Lite interleaved encoder and KDA | DINO with ResNet-50 backbone · COCO val2017 | AP: 50.7 -> 50.4 | Hurt |
| Replace DINO's encoder with the Lite interleaved encoder and KDA | DINO with Swin-T backbone · COCO val2017 | GFLOPs: 243 -> 159 | Helped |
| Replace the Deformable DETR encoder with Lite-Deformable DETR H3L1-(2+1)x3 | Deformable DETR, ResNet-50 · COCO val2017 | AP: 46.8 -> 46.7 | Hurt |
| Drop the fourth feature scale (ablation) | DINO, ResNet-50 · COCO val2017 | AP: 50.7 -> 48.2 | Hurt |
| Add key-aware deformable attention to the 3-scale DINO baseline (ablation) | DINO, ResNet-50 · COCO val2017 | AP: +0.8 AP | Helped |
| Add iterative high-level feature cross-scale fusion to the 3-scale DINO baseline (ablation) | DINO, ResNet-50 · COCO val2017 | AP: +0.8 AP | Helped |
| Add efficient low-level feature cross-scale fusion after high-level fusion (ablation) | DINO, ResNet-50, 3 scales · COCO val2017 | AP: +0.8 AP | Helped |
| Add KDA to the DINO 3-scale model with high- and low-level fusion (ablation) | DINO, ResNet-50, 3 scales · COCO val2017 | AP: +0.6 AP | Helped |
| Add high- and low-level feature fusion to the 2-scale DINO baseline (ablation) | DINO, ResNet-50 · COCO val2017 | AP: +4.0 AP | Helped |
| Add KDA to the 2-scale DINO model with high- and low-level fusion (ablation) | DINO, ResNet-50, 2 scales · COCO val2017 | AP: +0.7 AP | Helped |
| Use Lite interleaved updates instead of simply retaining two scales (ablation) | Deformable DETR, ResNet-50 · COCO val2017 | AP: 40.3 -> 45.8(+5.5) | Helped |
Closest papers by meaning
- Sparse DETR: Efficient End-to-End Object Detection with Learnable Sparsity (ICLR 2022 · similarity 0.91)
- DETRs Beat YOLOs on Real-time Object Detection (CVPR 2024 · similarity 0.88)
- CP-DETR: Concept Prompt Guide DETR Toward Stronger Universal Object Detection (AAAI 2025 · similarity 0.86)
- MS-DETR: Efficient DETR Training with Mixed Supervision (CVPR 2024 · similarity 0.86)
- KD-DETR: Knowledge Distillation for Detection Transformer with Consistent Distillation Points Sampling (CVPR 2024 · similarity 0.86)
- MI-DETR: An Object Detection Model with Multi-time Inquiries Mechanism (CVPR 2025 · 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.