bigSUR details

additional requirements

info on bigsur.

to get the bigsur optimisation and window detection to run:

1. install and license gurobi optimiser 9. ensure gurobi is on your library path (LD_LIBRARY_PATH=/path/to/gurobi/lib) before you start chordatlas.
2. if you want to detect features (doors, windows...), install nvidia-docker. tested on an 8gb nvidia card under ubuntu.
3. follow the bigsur data instructions below.

here are rough-and-ready videos showing mesh reconstruction, and working with panoramas.

data

several datasets are available here. we don't have a license to distribute the original data used in the paper, so the GIS data has been replaced by openstreetmap, and streetview photos have been updated.

workflow

1. unzip the data
2. start chordatlas
3. select: file, open..., then select the tweed.xml in the root of the unzipped dataset
4. select the layer panos, then click download to fetch the panoramas from google. watch the command line for progress. this will take up a bunch of disk space (5mb per image).
5. (the mesh and gis data may cover different areas. you can select the minimesh layer and select load all to show all the mesh data; and "hide all" to hide it.)
6. find a block surrounded by panoramas with mesh data; use the select tool, and right click on it to create a block mesh layer.
7. select block in the layer-list, and click render panoramas to create the 2d street-side images
8. select block in the layer-list, and click find image features to detect windows etc... with a CNN. this is slow, with limited feedback on the commandline.
9. select block in the layer-list, and click find profiles. wait for the profiles to become visible in 3d.
10. select the profiles layer and click optimize to run the optimization, and create the resulting mesh. cancelling the optimization in the pop-up will create the mesh using the current best-solution (if any).
11. the file -> export obj menu option will export all visible polygons