cmu-ci-lab / dual_pixel_defocus_estimation_deblurring Goto Github PK

Hi,

Would you please share the code for the calibration steps?

Thank you!

Hello, thank you for the great work.
It seems like the dataset in this repository is in raw format,
different from the images in https://imaging.cs.cmu.edu/dual_pixels/supplement/001/index.html.

After running the code, how do we compare it with the groundtruth deblurred image (which is not in raw format). Do you apply gamma correction??

Thank you.

Hello,

Can you explain why the depth of the blur kernel is 48?

Thank you!

The values of GT images exceeds the 2^14 and does not match the values of raw inputs. How should I use the pre-processed GT?

Hi,

I follow the code below

with PIL.Image.open(path_to_file) as f:
      image = onp.array(f) - 1024
      image[image < 0] = 0
      image = np.stack([np.float32(image)] * 1, axis=2) / (2 ** 14 - 1)
      # image = np.float32(image) / (2 ** 14 - 1)
    return image

but the output images are still darker than what shown on the site here. Could you please tell me why?

Hi, I am a bit confused about the 2 Hour reported time of the optimization in paper and the readme:
"Each optimization runs for 10,000 iterations with an Adam optimizer, and takes about 2 hours on an Nvidia Titan RTX GPU."

Does that mean for each DP pair, if one wants to get a defocus-map, one has to run the optimization for 2 hours? Or this could be done once for a Pixel device using the calibration pattern of discs, and then the inference should be very quick given the DP pair and original kernels?

Thanks,
Touqeer