DAB-DETR: Dynamic Anchor Boxes are Better Queries for DETR
arXiv:2201.12329
Abstract
We present in this paper a novel query formulation using dynamic anchor boxes for DETR (DEtection TRansformer) and offer a deeper understanding of the role of queries in DETR. This new formulation directly uses box coordinates as queries in Transformer decoders and dynamically updates them layer-by-layer. Using box coordinates not only helps using explicit positional priors to improve the query-to-feature similarity and eliminate the slow training convergence issue in DETR, but also allows us to modulate the positional attention map using the box width and height information. Such a design makes it clear that queries in DETR can be implemented as performing soft ROI pooling layer-by-layer in a cascade manner. As a result, it leads to the best performance on MS-COCO benchmark among the DETR-like detection models under the same setting, e.g., AP 45.7\% using ResNet50-DC5 as backbone trained in 50 epochs. We also conducted extensive experiments to confirm our analysis and verify the effectiveness of our methods. Code is available at \url{https://github.com/SlongLiu/DAB-DETR}.
arXiv commentAccepted to ICLR 2022
Cited by in corpus (10)
- ARS-DETR: Aspect Ratio-Sensitive Detection Transformer for Aerial Oriented Object Detection
- SpecDETR: A transformer-based hyperspectral point object detection network
- Mask2Map: Vectorized HD Map Construction Using Bird's Eye View Segmentation Masks
- MS-DETR: Natural Language Video Localization with Sampling Moment-Moment Interaction
- Focused Decoding Enables 3D Anatomical Detection by Transformers
- Table Detection for Visually Rich Document Images
- Parametric Primitive Analysis of CAD Sketches with Vision Transformer
- Dense Object Detection Based on De-homogenized Queries
- To Make Yourself Invisible with Adversarial Semantic Contours
- Holi-DETR: Holistic Fashion Item Detection Leveraging Contextual Information