Hi there, I was a student of parts 1 & 2, but I’m also the developer behind Silicon Valley’s “Not Hotdog” app, which tells you whether the object in front of you is a hotdog… Or a not hotdog.
I finally got a chance to write a blogpost about how exactly we built it, with lots of architecture details (and code!)
If you have any questions, I’m doing a small Q&A on Hacker News…
I wanted to thank Rachel & Jeremy so much for this course — I literally could not have built the app without them. I also want to thank fellow student Brad Kenstler for his awesome CLR library for Keras which I used to make this app!
Now this is too funny. I did NOT think it was an actual, working app–just figured it was some pre-made recording of some sort.
[SPOILERS] – Well, no major plot points, so shouldn’t be of much care.
I was pulling my hair out at how much novelty they were giving to image recognition and that they were trying to get students to create the labeled data. JUST USE IMAGENET!!! Haha.
Loved that they went as far as to specify that the folder name had to match the object, which is of course how keras does classes.
Anyways, very cool that it was a product of this class! Love it!
Awesome! Well done and thanks for the detailed blog post - the codepush stuff is super interesting!