For the past 4–5 days, I was playing with a Deep Learning-based Color Transfer algorithm. Then I discovered Darktable, and oh boy, I was quite surprised by how powerful this tool can be. The first image is the RAW image without any corrections and grading
Note: I am a total newbie with really basic knowledge of color correction and grading, so you will see a lot of errors in the image. Feel free to criticize, as it will help me understand what can be done to improve my skills
Welcome to the forum. I am probably…no definitely confused by what you mean by Color Transfer Algorithm and then the Color Transfer RGB module. I would be interested in understanding about these. Thanks
Hey, thank you for your war greeting, you are totally right, and it’s my bad for causing the confusion. What I meant by “Color Transfer Algorithm” was that I was working with a Deep Learning-based color matching algorithm. It generates a LUT from a reference image. Instead of calling it Color Match Algorithm, I mistakenly wrote Color Transfer, which caused the mix-up. Hope that clarifies things!
Thank you @Doug-Phoenix for pointing out that detail about default processing. I honestly thought it was completely unprocessed because I just converted the .CR2 file directly to JPEG, so that step must have happened during the conversion process.
And, Thanks a lot for sharing the guide and the resource links as well. I’ll definitely dive into those to get a better handle on how Darktable handles things
I see, interesting experiment! I can see that the trees in the background have clearly this blue-ish/green tint from the reference, but how should the model know what do with the orange? Interestingly, it looks like the saturation was just bumped to max, but that is I think not what the reference image looks like (although, the dark colors could be quite saturated)
Okay… I have to admit I don’t understand the process then for what do you need the reference image then? Especially in this case, why are reference and input image so different in their appearance? Maybe could you explain the whole workflow a bit better, i.e. what the intent of the reference is and how the images are fed into the DNN?
Soon I will write a detailed post about the entire workflow, as well as the DNN architecture, and i will post here, and will mention you as well you for the read-up
for what do you need the reference image then? Especially in this case, why are reference and input image so different in their appearance? Maybe could you explain the whole workflow a bit better, i.e. what the intent of the reference is and how the images are fed into the DNN?