albertotono
commented
3 years ago iou iou_voxels kl loss rec_error
class name
airplane,aeroplane,plane 0.598022 0.474756 0.0 24.711739 24.711739
bench 0.497740 0.522154 0.0 69.006543 69.006543
cabinet 0.751526 0.747332 0.0 226.335480 226.335480
car,auto,automobile,machine,motorcar 0.748585 0.784312 0.0 77.747527 77.747527
chair 0.535337 0.556275 0.0 259.577288 259.577288
display,video display 0.559103 0.593989 0.0 249.017495 249.017495
lamp 0.400733 0.378115 0.0 135.292487 135.292487
loudspeaker,speaker,speaker unit,loudspeaker sy... 0.678105 0.713246 0.0 555.596922 555.596922
rifle 0.479959 0.521721 0.0 21.929605 21.929605
sofa,couch,lounge 0.700020 0.717558 0.0 185.688441 185.688441
table 0.540937 0.551061 0.0 192.542849 192.542849
telephone,phone,telephone set 0.753615 0.745751 0.0 72.385115 72.385115
vessel,watercraft 0.561711 0.580268 0.0 56.858065 56.858065
mean 0.600415 0.606657 0.0 163.591504 163.591504
also why in the unconditional KL goes to 16?
After running the evaluation with onet_pretrained.yaml, I noticed that the Kullback–Leibler divergence is always 0,
How did you measure it? and where can I find more info on this evaluation? Why is 0? and what is the best way to interpret this result
Best Regard, and great work