search

XinJCheng / CSPN

Convolutional Spatial Propagation Network
496 stars 92 forks source link

Did you up-sampling the depth when eval? #23

Closed LINA-lln closed 5 years ago

LINA-lln commented 5 years ago

Hi Thank you very much for your contribution. I have a question when I read your paper. Did you up-sampling the depth to the original size when evaluation?

XinJCheng commented 5 years ago

thanks for your attention, As shown in the paper, the output size for the last layer in the network should be same with the input size.

LINA-lln commented 5 years ago

thanks for your attention, As shown in the paper, the output size for the last layer in the network should be same with the input size.

And the output depth is directly evaluated according to the metrics with no up-sampling to 480*640 in NYUDv2 ? the same as kitti

XinJCheng commented 5 years ago

We first downsampled to half and then center-cropped to 304x228 as input, same with our baseline

longyangqi commented 5 years ago

Thank you for your code！ according to your eval.py , you directly compute the metrics on the cropped output depth and gt (eg. 304*228 for nyu), without back to 640*480 ?