From what I gathered from multiple issues ( #263 #261 #162 ), The training json should look like this:
{ "prompt": "description of reference", "reference": "reference.jpg", "output": "groundtruth.jpg" }
but I want to provide the training data in this format:
{ "prompt": "description of the reference image", "mask": "blackandwhitemask.jpg", "reference": "reference.jpg", "input": "originalimage.jpg" "output": "inpaintedoutputusingthemask.jpg" }
is it possible to create a dataset like this ? is there a better way ? I think I should focus on tutorial_train_plus.py, would I need to change anything so that I can provide the mask as well for inpainting ? How big the training dataset should be If I want to fine tune it ? Just to clear, I am not training it on faces.
From what I gathered from multiple issues ( #263 #261 #162 ), The training json should look like this:
{ "prompt": "description of reference", "reference": "reference.jpg", "output": "groundtruth.jpg" }
but I want to provide the training data in this format:
{ "mask": "blackandwhitemask.jpg", "reference": "reference.jpg", "input": "originalimage.jpg" "output": "inpaintedoutputusingthemask.jpg" }
or
{ "prompt": "description of the reference image", "mask": "blackandwhitemask.jpg", "reference": "reference.jpg", "input": "originalimage.jpg" "output": "inpaintedoutputusingthemask.jpg" }
is it possible to create a dataset like this ? is there a better way ? I think I should focus on tutorial_train_plus.py, would I need to change anything so that I can provide the mask as well for inpainting ? How big the training dataset should be If I want to fine tune it ? Just to clear, I am not training it on faces.