Can we start bfgs from a previous state?

I have tested my model for 10, 50, 150 iterations, now I want to try 200+ iterations to get better results. It seems a waste to always start from the beginning. is there a way for us to start bfgs from a previous states?

below are my experiments with a starry night style image.
my content image is from the woody allen movei “midnight in paris”

2 Likes

I think you can look at this paper that talks about color preservation. It would help you get closer to the goal.
https://arxiv.org/abs/1606.05897

1 Like

Sure, just feed in your current version of x, rather than initializing x to random.

1 Like