Hi,
Thanks for sharing the code. I have a question about the evaluation method in MATLAB code.
In the evaluateNYU.m, line 37-49, given a 480640 image, the code seems
(1) resize it to 240320,
(2) do center-croping to 228304,
(3) do a prediction which I assume the output should also be 228304 corresponding the center area (I cannot make the matlab code running because of the package issue),
(4) perform bilinear upsamping to make it 480*640 depthmap and compare with ground truth.
Please correct me if my understanding is wrong.
If my understanding is correct, then the prediction depth can only cover the center area (228*304), and after upsampling, it seems the depth map will not be able to align the ground truth (in pixel level) any more. It means the prediction depth in the image edge area (at least) is actually not a prediction based on any corresponding input RGB image pixel.
Are you intentionally do this, or it is a bug? I am confused about the intention behind the evaluation process. Thanks in advance!
Hi, Thanks for sharing the code. I have a question about the evaluation method in MATLAB code. In the evaluateNYU.m, line 37-49, given a 480640 image, the code seems (1) resize it to 240320,
(2) do center-croping to 228304,
(3) do a prediction which I assume the output should also be 228304 corresponding the center area (I cannot make the matlab code running because of the package issue), (4) perform bilinear upsamping to make it 480*640 depthmap and compare with ground truth. Please correct me if my understanding is wrong.
If my understanding is correct, then the prediction depth can only cover the center area (228*304), and after upsampling, it seems the depth map will not be able to align the ground truth (in pixel level) any more. It means the prediction depth in the image edge area (at least) is actually not a prediction based on any corresponding input RGB image pixel.
Are you intentionally do this, or it is a bug? I am confused about the intention behind the evaluation process. Thanks in advance!