It won't be long and you'll be stringing together a bunch of Neural Networks with some glue in the middle where each of the nets performs some task.
One net to crop the image, one net to recognize it, another to synthesize the words to describe it and presto: a new application.
This is just a data pipeline . Multiple tools support what you describe . Or you could do this as a batch job / use spark to perform the task as a stream.
Yeah I'd envisaged this for local second hand websites to better crop and auto categorise images.
This is how a lot of things work already. Neural nets are bad at strict logic. If you give one net one of two tasks, depending on a state, it takes a lot of training before the learned tasks separate from each other. Ideally, the aspects unique to each task would be completely separate and anything common to both tasks would occur once in the net and get used for both. In practice, it increases training requirements by so much that you are better off creating 2 nets and logically separating them with your own code or even creating a 3rd net to detect which state you are in.
What you describe is how I imagine the first recognizably intelligent AI system being built -- a set of communicating independent modules all running at the same time.
this is already how it works
Cool! For any javascript devs this looks like a capable library that attempts the same functionality and looks like it has pretty impressive results: https://github.com/jwagner/smartcrop.js/
Thank you for that share, this is something I want after having read the article for a project I'm just starting on.
Cool tech, but seems like over-engineering. As far as I can tell, Facebook just crops to the center and top of images which works pretty well. Meanwhile Twitter has spent years cropping images awkwardly based on trying to find the relevant regions, and now is doubling down on trying to find relevant regions...
This is really cool! The SmartyCrop service does a really good job of intelligently croppping images as well. https://smartycrop.com/
I have seem good results in automatic cropping by just giving a higher value to areas of high entropy and to faces. I wonder how this compares.
They specifically mention that their previous face detection-based algorithm didn't work as well as they'd like. E.g. for animal faces.
so is there some open source library or something that I can try, or just a paper announcing this new feature at twitter?
Is that why all my pasted images show up as blank?