fabfabBesnard
commented
3 years ago Test on AWS with db_eval_1 (12 plants); config file: ~/configs/ml_pipe_fab_fixAI.toml. analysis date: 2020/10/23 The good: -all plants pass the reconstruction -segmented point clouds are overall correct (excpet for GT2: pcd completely wrong, with a "plane" above the plant)
The bad:
-organs fruit points are often wrong. Many fruits are missing (although they are in the segmented point cloud), often at the base of the plant. For one scan, they are transformed into a huge and ugly shape that make no sense (cf picture). Corresponding .ply in the ClusteredMesh folder are also like this, so the pb could come from the computing of the clustered mesh
e.g.: 2018-11-27_15-13-03 (plant Col0_12_10_2018_B)
Type of errors with a apparently good segmented pcd: -the fruit is missed (no organ highlight/no fruit point), especially at the base of the plant -permutation in the order of fruits (especially at the tip of the stem)
Some errors can be attributed to errors of segmentation: -tip of leaf segmented as fruit -fruit segmented as leaf
coucou !
It's not clear what the result of commit 5ba46f0 + 0896517 for geom pipe: you updated the computation of angles then you put back the previous version ? Nobody knows about the content of the version (except you !). So please be clear on the effects (desired or undesired) that you modified by each commit. Could you comment a clear message about the status of angles/internodes sequences obtained by this fix ? Like: are pbs still present ? / is there a unique direction set for an entire plant ? Are all angles expressed according to this direction ? How is this direction set ?