Absolutely!
With visual content being predicted for popularity, a model can be made which produces quality content based on the training data of existing popular content from different sources.
A neural network has been trained to differentiate between popular and non-popular images and give like scores. This network was optimised to understand the features that made a photo popular on social media. Part of the work was inspired by Karpathy's blog on selfie (http://karpathy.github.io/2015/10/25/selfie/). However, our algorithm works on different type of images and not just selfies.
We have trained our model on public photos which were popular/trending at some point of time. We fed these images to a deep CNN (Convolutional Neural Network) which started to recognise features that made photos popular. What we realise in the process that these features do change with time so we have added a temporal component to our training set to ensure our model is relevant.
It is currently optimised more for human photos and travel images so maybe in our next iteration, we can predict pets photos as well.
A large majority of human errors come from fine-grained categories(such as correctly identifying two similar cat species) and class unawareness. I would recommend this article by Andrej Karpathy, where he talks about his learning from competing against GoogLeNet: http://karpathy.github.io/2014/09/02/what-i-learned-from-com...
Yes! Facebook's Densenet won the best paper award in CVPR this year.
I would recommend PyTorch framework to code these in as it extends the numpy, scipy ecosystem and is simpler to use.
What are your thoughts?