Finding Deep Learning Models To Try2020-10-30
This is not a definite guide to finding interesting models to try. More of a note for future-me.
First of all, I think it's important to have some kind of guiding curiosity. Having a source of inspiration is great!
You can check out recent inspiring developments and see what looks cool. Of course, just giving a nice tool a try instead of trying to run it yourself (or build it yourself) is a completely acceptable initial step! See if the application resonates enough that more than a bit of tinkering comes out of it. (For example, playing around with artbreeder or some "this X does not exist" type of site is a valid thing to do!)
If you have a project which interest you, and you'd like to replicate the functionality, look for an about section or related blog posts. Sometimes, there's an open-source model mentioned or directly linked.
If you're into a certain topic (like GAN for example), look for an overview of approaches - either as a GitHub repository, an "awesome X" list or an article (like this one about GAN architectures). You'll have a term to Google for once you know what exactly you're looking for.
GitHub or Twitter are additional sources. When a new paper is published, it's not unusual to have a some kind of model or POC code published as well. Sometimes those repositories are framework-specific. (For example, there is a PyTorch Image Models repo).
Finally, there are framework-specific model hubs. PyTorch Hub, the TensorFlow Hub and the TensorFlow Model Garden for example.
UPDATE: There's also Caffe (what a presentation) and their Model Zoo
UPDATE 2: Deeplearning-models on GitHub.
UPDATE 3: Papers with code and deep learning gallery.