Questions about how to generate custom depth

[jiayaozhang]

Hi, thanks for your interesting work!

Can I ask some questions?

• I tries to inference with your DCTNet and after I tried processing_testsets.py and get the RGBD_01_output_depth-depthLrUp.npy how could I transfer the npy file into png file.

• For testing, if it also supports the custom images? should I need to re-train the model or I can directly use the pretrain model to generate the custom depth images by DCTNet

[jiayaozhang]

for example, Middlebury_Afterprocessing_4x, these would generate three datasets, DepthHR, DepthLrUp, RGB in DatasetsAfterprocessing.

And in each DepthHR, DepthLrUp, RGB it would generate the corresponding npy file, but which one is the npy that you showed in the paper? And how I can tranfer the npy into png file

Many thanks!

[jiayaozhang]

I found that only DepthHR is working.
DepthLrUp and RGB didn't generate decent images

And attached the method for generating image from npy

import glob
files=glob.glob('images/*.npy')
for file in files:
 png_name=file.replace('npy','png')
 a=np.load(file)
 b= a[0]
cv2.imwrite(png_name,b)

[jiayaozhang]

One last question is why 4x, 8x and 16x images are the same resolutions for different datasets, thanks for help!