Copycat node is a machine learning algorithm introduced in 2020 in nuke, it is used to train with a set of images and generate in-between frames with data which matches the given intervals of frame. the data keeps getting better when we feed more data and the output becomes more accurate.
Examle:
We can use the copycat to clean-up the plate by training the node with intermediate cleaned up frames and letting the machine learning to create the in-between frames.
- Create the Copycat node and set the directory
- Copycat node should work on linear colorspace input.
- Adjust epochs/steps of calculations as per the shot requirement / Think about epochs like sampling in 3D/ MB.
- More the epochs more the time to train and also the PC might be unusable during the calculations.
- In the current version of copycat we can pause the training process to check the results.
- In the advance settings having large model size might take a lot time to calculate so need to be carful on using these settings.
- We can use any previous training data and use it as a checkpoint to aid the machine learning.
- We can use the same principle to many things like the roto