This would be a good idea when I am shopping for someone I am not quite sure about. Getting ideas on the spot could be a good inspiration especially with the holiday season coming up
Thanks for all the feedback. After the successful 2 week beta, we are now launching the next version this week all new features and integration with social media. I may make another ph post for that
We are actually more interested in how to integrate with Instagram than to create an app.
If you are curious about our technology I can say that is mostly composed by two pieces: a deep learning descriptor called deepDNA and the street to shop solver that uses deepDNA.
DeepDNA is every efficient in computation and merges image features and text features (word embeddings) so contains a lot of information.
It captures incredible semantics, not just only visual features. The recommendations are mindblowing. But our biggest technology improvement is probably how we work with natural images.
Mostly all the visual technologies out there are doing the following pipeline: Product extraction and reverse image search.
We started doing this also. We developed technology based on top of state-of-the-art segmentation / object detection papers...
However my opinion is that this pipeline is very hard to make it work for every single scenario (partial products to full human images).
It can work apparently well, but again, very difficult to make it work always.
We also want to cover every space, including home design and annotate images does not sound even realistic to me.
So we had to think out of the box to not doing what we consider a mistake.
We still use RPN's but more as an attention model. I would say that our approach to solve the street to shop problem is more RNN than CNN.
Believe me, we are working hard to make that game changer ;)
DoodleLens