Hi, I am new to gmic and its CLI. I am trying to fill a specific area in several thousand images using gmic. I successfully done it in one image using following command:
gmic a.jpg 100%,100% polygon[-1] 5,345,4095,475,3976,700,4044,901,3875,1495,4095,1,255,255,255 +inpaint_pde[0] [1]
This mask is present only in the lowest part of the image. If I try same in gimp, its processed instantly using gmic gimp qt (GUI).
I have thousands of images to process and each with same size of around 3000 x 4000. Each image with above command is taking long time to process. Can any one suggest a solution to that. I read somewhere that we can divide the image into parts, but I wasn’t able to find that in G’MIC reference documentation.
The PDE-based inpainting algorithm does not require a lot of context to work, which means you can indeed extract the image area corresponding to the region you want to inpaint, surrounded by a bit of known pixels, then apply the inpainting algorithm on this region, and paste the result back to the original image.
In you case, something like:
gmic a.jpg 100%,100% polygon[-1] 5,345,4095,475,3976,700,4044,901,3875,1495,4095,1,255,255,255 +z 330,3800,1300,100% inpaint_pde[2] [3] j[0] [2],330,3800 k[0]
should work.
1 Like
Wow thats fast… thank you so much, thats exactly what I was looking for… A little more help I appreciate if you can describe a little whats happening here, its a little complicated, I didnt really understand the “z [3] j k” …
The command z
is a shortcut for command crop
, so +z ...
extracts some small region from the input image and the mask. After the crop, your list has 4 images :
-
[0]
= Fullres color image
-
[1]
= Fullres binary mask
-
[2]
= Cropped region from color image.
-
[3]
= Cropped region from binary mask.
Now, we apply inpaint_pde[2] [3]
to reconstruct cropped color image [2]
using the inpainting mask [3]
, and we finally redraw the reconstructed region [2]
into the fullres color image [0]
(command j
, aka image
). Latest command k
(aka keep
) only keeps the fullres reconstructed color image, which is the one that is interesting.
1 Like
Thanks for your prompt reply, appreciate that