Currently, I am working on utilizing AnimateDiff to generate a video from a given image, such as the stainless cup in the GIF provided below. I intend to modify the background color (fog in the GIF) using the 'Color Transfer' feature within this project.
The expected behavior of 'Color Transfer' is to take a source video and a target video composed of an image sequence and a single mask image (or a mask image sequence) as inputs and produce a color-transferred image sequence as the output.
However, the output of 'Color Transfer' currently results in a single color-transferred image.
My suspicion is that this problem may stem from the ColorXfer2 method within the Color_Correction class, as it seems to eliminate the batch dimension of the mask tensor.
I'm uncertain about the clever solution to this, but I believe it could potentially be resolved by preserving the batch dimension.
zae-park
commented
1 year ago
I'm pleased to have discovered this project.
Currently, I am working on utilizing AnimateDiff to generate a video from a given image, such as the stainless cup in the GIF provided below. I intend to modify the background color (fog in the GIF) using the 'Color Transfer' feature within this project.
The expected behavior of 'Color Transfer' is to take a source video and a target video composed of an image sequence and a single mask image (or a mask image sequence) as inputs and produce a color-transferred image sequence as the output.
However, the output of 'Color Transfer' currently results in a single color-transferred image.
My suspicion is that this problem may stem from the
ColorXfer2method within theColor_Correctionclass, as it seems to eliminate the batch dimension of the mask tensor.I'm uncertain about the clever solution to this, but I believe it could potentially be resolved by preserving the batch dimension.