search

nianticlabs / stereo-from-mono

[ECCV 2020] Learning stereo from single images using monocular depth estimation networks
Other
398 stars 54 forks source link

Metrics used for evaluation #3

Closed sieme97 closed 4 years ago

sieme97 commented 4 years ago

Hello, your paper is quite interesting. Did you evaluate the proposed method on commonly used metrics such as absolute relative error, RMSE etc?

JamieWatson683 commented 4 years ago

Hi, thanks for your interest, and sorry. for the slow reply.

Unfortunately no, we did not - we instead opted to use the metrics generally used by the deep stereo community (i.e. metrics on disparity not depth).

I've just pushed an initial code release, so you could adapt the inference code to additionally compute RMSE relatively easily, just make sure to convert all disparities (both predictions and ground truth labels) to depths before computing the scores.

Thanks a lot, Jamie