When a predicted instance intersecting with ignored category such as wall and floor(e.g. labels==1 or 2), why should add the num_ignore to increase the proportion_ignore and avoid the FP(false positive). In my opinion, this will cause FP to be inaccurate. Thank you, could you please help me to understand this?

[RongkunYang]

https://github.com/ScanNet/ScanNet/blob/3e5726500896748521a6ceb81271b0f5b2c0e7d2/BenchmarkScripts/3d_evaluation/evaluate_semantic_instance.py#L156