TaPaD: An Instance-Level Image Dataset of Deteriorated Tactile Paving
收藏资源简介:
TaPaD (Deteriorated Tactile Paving) is an instance-level image dataset for tactile paving detection and deterioration assessment. The dataset consists of images collected using a bicycle-mounted compact camera and smartphones in public environments, primarily in Okayama Prefecture, Japan, with a small number of samples collected outside the prefecture and overseas. Each distinguishable tactile paving block is manually annotated using bounding boxes. The dataset includes tactile paving type labels (guiding and warning), multiple deterioration labels (e.g., cracked, dirty, faded, melting, and worn-out), GPS metadata when available, and accompanying metadata. The archive also provides the original annotation format, annotation specifications, annotation guidelines, and conversion scripts for generating YOLO and multi-hot formats. This dataset is intended for research on computer vision, accessibility, infrastructure inspection, and assistive technologies.



