Line Segment Detection Using Transformers without Edges
Line Segment Detection Using Transformers without Edges
复制标题
DOI：
10.1109/cvpr46437.2021.00424
复制
发表时间：
2021-01
期刊：
影响因子：
--
通讯作者：
Yifan Xu;Weijian Xu;David Cheung;Z. Tu
中科院分区：
文献类型：
--
作者：
Yifan Xu;Weijian Xu;David Cheung;Z. Tu
在本文中，我们提出了一个联合端到端的线段检测算法，使用变压器是后处理和几何学引导的中间处理（边缘/交界处/区域检测）免费。我们的方法，命名为线段TRansformers（LETR），利用集成标记化查询，自我关注机制，并通过跳过标准的启发式设计的边缘元素检测和感知分组过程中的变形金刚内的编码-解码策略。我们为Transformers配备了多尺度编码器/解码器策略，以在直接端点距离损失的情况下执行细粒度的线段检测。该损失项特别适合于检测几何结构，例如无法方便地用标准边界框表示法表示的线段。变形金刚学会了通过自我关注的层次来逐渐细化线段。在我们的实验中，我们展示了Wireframe和YorkUrban基准测试的最新结果。
In this paper, we present a joint end-to-end line segment detection algorithm using Transformers that is post-processing and heuristics-guided intermediate processing (edge/junction/region detection) free. Our method, named LinE segment TRansformers (LETR), takes advantages of having integrated tokenized queries, a self-attention mechanism, and encoding-decoding strategy within Transformers by skipping standard heuristic designs for the edge element detection and perceptual grouping processes. We equip Transformers with a multi-scale encoder/decoder strategy to perform fine-grained line segment detection under a direct endpoint distance loss. This loss term is particularly suitable for detecting geometric structures such as line segments that are not conveniently represented by the standard bounding box representations. The Transformers learn to gradually refine line segments through layers of self-attention. In our experiments, we show state-of-the-art results on Wireframe and YorkUrban benchmarks.