So I have a few film styles I like to emulate, some colour, some b/w. I have sample photos for reference for consistency.
When I try and drop a b/w image onto a colour photo (and all digital files are really colour) the ML is still giving me a colour output.
You'd think an all-b/w image as reference would be "learned" by the ML to have a b/w output.
Perhaps a way to indicate that the reference image is b/w.
I've tried this over and over and the ML still hasn't "learned" the difference. Since these are the two main genres of image, Id think it would be better.
I do like the concept of ML for consistent filter application. This could be really handy for portfolios. Not quite there yet.
Thanks.
Machine Learning Quandary
2019-07-29 14:39:35
The ML algorithm doesn't learn from what you do with your photos, it is a 'trained' algorithm that, in theory, for the same two photos, should always provide exactly identical results. In addition, we actually didn't train our algorithm specifically for identifying black and white photos, its ability to do so is more a byproduct of the algorithm and dataset. However, this is definitely an interesting feature, perhaps for a future update.