It started with a joke.
Not my joke, but a good one.
To some, this is not a joke. It is a convoluted neural network similar to the TV show Silicon Valley’s Not Hotdog app
The Vision
The aim of this project is to deploy a complete AI solution using crowd-sourced data. Eventually, it consists of three components:
- A Web App to crowd source labelling training data
- Training of a AI/Machine Learning model. The first iteration will likely be trained on an NVIDIA GTX 960M
- An Inference App to use the model, and allow users to contribute new images for training
Cycle round, train more accurate models with more training observations, more parts etc..
To date, part one is in “beta” testing on this site. Part two is being dabbled with offline. Part three is still a pipe-line dream.
Does it work?
Well no not really. Not yet at least. Initially the model is based on images of my Lego and pieces I have tagged. The more observations the model can be trained with then the more accurate the finds can be.
If you’d like to help improve this project then please do head on over to the tag-bricks part of this site to create more observations.
