Scalable Label-efficient Footpath Network Generation Using Remote Sensing Data and Self-supervised Learning
PyTorch implementation and models for footpath segmentation and GIS layer generation. For details, please see our paper.
Our pipeline can be divided into two main parts, footpath segmentation and GIS layer generation.
The first part contains two training processes involving the construction of two models, self-supervised pre-training (DINO-MC) and footpath segmentation fine-tuning (U-Net).
The code of the self-supervised pre-training can be found at DINO-MC.
We provide code of the second part which converts the model output mask to a GIS layer. The implementation references the Tile2Net.