Daniel from Mapbox here. Happy to answer questions or talk through design decisions. Interested to hear your feedback.<p>We are mostly focusing on making the process accessible to a broader audience in the geo space, building a solid production-ready end-to-end project.<p>There are more resources and a step-by-step guide for running on openly available drone imagery in Tanzania:<p><a href="https://www.openstreetmap.org/user/daniel-j-h/diary/44145" rel="nofollow">https://www.openstreetmap.org/user/daniel-j-h/diary/44145</a>
<a href="https://www.openstreetmap.org/user/daniel-j-h/diary/44321" rel="nofollow">https://www.openstreetmap.org/user/daniel-j-h/diary/44321</a>
Have you released pre-trained models?<p>It would be pretty useful if you did, even as a just basis for transfer learning.<p>Also a description of the model that is used? I assume this is the code[1], which references <a href="https://arxiv.org/abs/1806.00844" rel="nofollow">https://arxiv.org/abs/1806.00844</a>, but the code doesn't seem to use WideResnet (although I really know Keras much better than PyTorch so I'm probably missing something.<p>[1] <a href="https://github.com/mapbox/robosat/blob/master/robosat/unet.py" rel="nofollow">https://github.com/mapbox/robosat/blob/master/robosat/unet.p...</a>
Interestingly, we've taken the same approach to process historical document (like 18th Venetian manuscripts).<p>We even use a Unet architecture with a pretrained resnet50 encoder, and some postprocessing to go from prob maps to polygons, like this project does. Of course, we are much more limited than what you propose, but it is reassuring our side project took the same course as what bugger entities do.<p><a href="https://dhlab-epfl.github.io/dhSegment/" rel="nofollow">https://dhlab-epfl.github.io/dhSegment/</a>
Hi Daniel, in your experience, what features is this model most suited for and with what granularity of imagery? For example, buildings/roads with landsat(30m)? Cars with 30cm resolution imagery?