anton_foy
23rd September 2024, 16:55
Someone here who has knowledge of how to train pix2pix with my own dataset of souce and target images but without training the spatial factors? The goal is to train it to predict color corrections from learning the corrections/changes from source to target to apply on any chosen input images. Why I do not want the spatial aspect is because I want it to adjust colors and brightness in the way a 3D LUT works, later I may add this as a feature of the model to output .cube files.
I am open for another model there is any suitable ones for this type of work. Thanks!
Edit: during the training phase I guess it can use spatial data to eg. detect objects etc. if it will help it understand the color corrections better, just not apply any spatial aspects during inference.
I am open for another model there is any suitable ones for this type of work. Thanks!
Edit: during the training phase I guess it can use spatial data to eg. detect objects etc. if it will help it understand the color corrections better, just not apply any spatial aspects during inference.