search

dimitrisPs / msdesis

MSDESIS: Multi-task stereo disparity estimation and surgical instrument segmentation
MIT License
22 stars 3 forks source link

About calculate Depth_MAE? #2

Open Tobyzai opened 4 months ago

Tobyzai commented 4 months ago

Hi author, I would like to ask when calculating depth_mae, why is the predicted depth value less than or equal to zero set to 40. what is the purpose of doing this?

’evaluation_scripts/evaluate_scared.py‘：

mae
dimitrisPs commented 4 months ago

Hi there,

This code was adapted from the evaluation scripts used during the endovis SCARED challenge. The idea is that if for whatever reason your model predicts zero or negative depth, a fixed penalty is applied to the corresponding pixel. By doing so, the challenge organisers were able to penalize methods that didn't produce dense depthmaps or depthmaps measuring distance behind the camera.