FootpathVision

The FootpathVision dataset is specifically designed for detecting sidewalk encroachment in dense urban environments using deep learning models. The dataset comprises 1,238 images collected from Dhaka, Bangladesh, consisting of manually captured images using mobile phone cameras and different image frames extracted from Creative Commons-licensed YouTube videos of Dhaka city. Images are divided into two classes: encroached footpaths and clear footpaths. Encroachments in the dataset include illegal stalls (vendor structures, pushcarts, makeshift shops, furniture) and illegal parking. The dataset is organized into folders based on encroachment status (encroached/non-encroached), with separate subfolders distinguishing between our manually captured images and YouTube-sourced frames.

Please cite our research paper that benchmarks this dataset using several deep learning models.