NVlabs / splatnet

SPLATNet: Sparse Lattice Networks for Point Cloud Processing (CVPR2018)

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

eval bug

jason718 opened this issue · comments

Shouldn't accuracy= TPs / (TPs+FPs)? What computed in below two lines seem to be recall.

scores['class_accuracy'] = [TPs[i] / (TPs[i] + FNs[i]) for i in range(len(labels))]

class_acc = [(tp[i] + eps) / (tp[i] + fn[i] + eps) for i in range(len(labels))]

This is how "per-class" accuracy is typically defined for this dataset. It can be considered as "accuracy among points belonging class i"