Research map / Detection and segmentation / CNN features
ViDT: An Efficient and Effective Fully Transformer-based Object Detector
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 the standard detector body and neck encoder with Swin plus RAM and an encoder-free decoder neck | ViDT with Swin-base · COCO2017 val | FPS: 4.8 ~~(5.4)~~ -> 9.0 (11.6) | Helped |
| Add neck decoder, auxiliary decoding loss, and iterative box refinement (ablation) | ViDT with Swin-nano · COCO2017 val | AP: 28.7 -> 40.4 | Helped |
| Use sinusoidal spatial encoding before projection rather than no spatial encoding (ablation) | ViDT without neck, Swin-nano · COCO2017 val | AP: 23.7 -> 28.7 | Helped |
| Replace pre-addition sinusoidal encoding with pre-addition learnable encoding (ablation) | ViDT without neck, Swin-nano · COCO2017 val | AP: 28.7 -> 27.4 | Hurt |
| Restrict cross-attention to the last stage rather than enabling it at stages 1–4 (ablation) | ViDT without neck, Swin-nano · COCO2017 val | FPS: 21.8 -> 36.5 | Helped |
| Add auxiliary decoding loss and iterative box refinement to the body without a neck (ablation) | ViDT without neck, Swin-nano · COCO2017 val | AP: 28.7 -> 22.9 | Hurt |
| Distill from a larger ViDT using token matching (ablation) | ViDT Swin-nano student · COCO2017 val | AP: 40.4 -> 41.5; 40.4 -> 41.9 | Helped |
| Drop two decoder layers at inference (ablation) | ViDT Swin-nano · COCO2017 val | FPS: 20.0 -> 22.3 | Helped |
| Remove the neck encoder (ablation) | Deformable DETR with Swin-nano · COCO2017 val | AP: 43.1 -> 34.0 | Hurt |
| Add a neck decoder to YOLOS (ablation) | YOLOS with DeiT-tiny · COCO2017 val | AP: 30.4 -> 38.1 | Helped |
| Add a neck encoder (ablation) | ViDT with Swin-nano · COCO2017 val | AP: 40.4 -> 46.1 | Helped |
| Add a neck decoder to YOLOS (ablation) | YOLOS with DeiT-tiny · COCO2017 val | FPS: 28.1 -> 17.1 | Hurt |
Closest papers by meaning
- A Transformer-Based Object Detector with Coarse-Fine Crossing Representations (NeurIPS 2022 · similarity 0.90)
- Training Object Detectors From Scratch: An Empirical Study in the Era of Vision Transformer (CVPR 2022 · similarity 0.89)
- An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale (ICLR 2021 · similarity 0.88)
- DESTR: Object Detection With Split Transformer (CVPR 2022 · similarity 0.88)
- TGBFormer: Transformer-GraphFormer Blender Network for Video Object Detection (AAAI 2025 · similarity 0.87)
- DECO: Unleashing the Potential of ConvNets for Query-based Detection and Segmentation (ICLR 2025 · 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.