Early Copycat Experimentation | Smart Vector Driven Colours
I went into this with fewer expectation than usual, but I was very interested in seeing what CopyCat would do if it has to find a pattern in something as ever changing as the the vector channels of a Smart Vector. I believe what ended up happening was the Neural Network took the general average colours and used them, while somewhat discarding frames that were unusual, which would make sense has I had a fairly high error average in the solve graph by the end of the training.
CopyCat Settings:
Epochs: 40000
Initial Weights: None
Model Size: Small
Batch Size: 8 (Auto)
Crop Size: 128
Checkpoint Intervals: 1000
Contact Sheet Interval: 100
Training Length: 6 Hours
RTX 2060 | i7-8700 | 32GB 2666MHz
The .cat(nc) files took up about 1.34GB after all the training finished.
Done in NukeX 13.0v4 in the AIR Copy Cat ML node.