search

IGNF / myria3d

Myria3D: Aerial Lidar HD Semantic Segmentation with Deep Learning
https://ignf.github.io/myria3d/
BSD 3-Clause "New" or "Revised" License
151 stars 20 forks source link

Dev: allow inference for smallest clouds possible + addresses many smaller issues. #81

Closed CharlesGaydon closed 11 months ago

CharlesGaydon commented 11 months ago

Goal : accept patches of data with the minimal number of nodes i.e. 1 node. Until now we would disregard patches with less than 50 nodes (points). But the model architecture is designed to accept such cases, which is useful to avoid null predictions in e.g. water areas.

There was an issue with the standardization of intensity / average colors when num_nodes was equal to 1 (then : std is undefined). This is now fixed and any any patch with at least 1 node is accepted and can have an inference.

Empty patches are still ignored.

This should fix https://github.com/IGNF/myria3d/issues/79

Additionnally this branch turned into a dev branch with other additions:

CharlesGaydon commented 11 months ago

rebased pour ne garder que le changement de la minor version.

CharlesGaydon commented 11 months ago

Error due to docker mount. I deleted the image and ran the tests again...

CharlesGaydon commented 11 months ago

Simply a store mount error due to a bad network (I cannot ping store.ign.fr). I unmounted the store and will need to mount it back when the connection is back on.

CharlesGaydon commented 11 months ago

Last commit fixes https://github.com/IGNF/myria3d/issues/60

CharlesGaydon commented 11 months ago

last commit fixes https://github.com/IGNF/myria3d/issues/68

CharlesGaydon commented 11 months ago

Last commit fixes: https://github.com/IGNF/myria3d/issues/69

CharlesGaydon commented 11 months ago

Last commit addresses https://github.com/IGNF/myria3d/issues/75

CharlesGaydon commented 11 months ago

@leavauchier J'ai rétabli les tests unitaires. Le store est toujours inacessible mais le reste des tests passe.