Every now and then I see people ask for projects for Machine Learning classes so I thought I'd pass on an idea I had but will not have time to work on.
I think it would be really interesting to see someone attempt to remove watermarks from images using an autoencoder. In many ways this can be see as similar to how a denoising autoencoder works, but effectively the noise would be the watermark. It would be very easy to generate training data and it would be effectively infinitely available.
Just thought I'd throw this out there for anyone interested especially given the popularity of neural nets right now.
[link][4 comments]