Deep-homography-estimation-Pytorch icon indicating copy to clipboard operation
Deep-homography-estimation-Pytorch copied to clipboard

Published 20 hours ago • verified publisher icon mazenmel

How's the performance?

Open eglxiang opened this issue 5 years ago • 2 comments

Thanks for re-implement the model. How's the performance? Consistent with claimed in the paper?

eglxiang avatar Apr 27 '19 05:04 eglxiang

Hi, We did not use the same metric described in the paper so I can not compare with the paper, we have calculated the MSE between the prediction and GT matrix.

mazenmel avatar May 11 '19 20:05 mazenmel

Hi, We did not use the same metric described in the paper so I can not compare with the paper, we have calculated the MSE between the prediction and GT matrix.

Hi, I've tested the regression homography performance using https://github.com/fjbriones/deep-homography, the MACE is about 48~50 which is significantly worse than that claimed in the paper. And after I implemented this work by myself in TF, I got the same result. Can u tell me the MSE in your experiment? Since the MACE is 9.2 for regression homography net in the paper, the corresponding MSE between the prediction and the GT should be about 42. Can you tell me the result of your experiment for reference?

nie-lang avatar Jul 24 '20 12:07 nie-lang