Joining the Coursera’s Machine Learning club

Today I completed what seems to be an inevitable step for whoever is interested in Machine Learning: the Machine Learning course by Stanford Prof. Andrew Ng on Coursera. 

It has been a really enjoyable, interesting and fascinating course, that brought me to wake up earlier in the morning and have lunch on my own, all to find the time to follow the course and study.

I highly recommend this course to all the people who want to get a solid basis on Machine Learning.

Thank you Prof. Andrew Ng, Coursera, and Stanford University for making this course free for everyone.

AI Projector: quickly prototype your computer vision software

Recently I have taken part in an hackathon focused on AI and Computer Vision organized by Atrovate. It has been a wonderful and enriching experience.

Something struck me though.

Of the many amazing projects presented at the end of the hackathon, only a very few of them had a proper UI and the presentation was mostly based on slides.

In some cases, this was due to the nature of the project itself.

Yet, for many projects a UI allowing interacting with the computer vision model would have provided a much richer experience.

The reason was clear: the focus of the hackathon had been put on the AI core, not the UI. This makes sense: the innovation of the project lies in its AI after all.

Another reason: not all the machine learning engineers have UI development experience.

I came out of this hackathon with an idea: to build a pre-baked UI that would allow machine learning engineers quickly prototype computer vision software, by focusing on its core, the AI model.

Today I’m releasing a first version of this project: AI Projector.

AI Projector is web-based and simple to use. You can just download the files in this repository, customize the settings in aiprojector.js and you’re ready to go.

AI Projector opens a video stream from your webcam or any video recording device you have, samples its video frames and send them to your classification model.

Frames that were labeled (i.e. the model returned at least one label with a probability >= threshold), are shown with their labels, and you can easily save them (depending on your browser, right-click + save image).

Once your model is ready, you can build your own UI or use AI Projector as a starting point for a web-based UI (AI Projector is released under the much permissive MIT license).

I’m playing with it and it feels like a good way to speed up experimenting with computer vision. 🙂

Hopefully AI Projector will come handy in many computer vision scenarios.

ArkAngel wins the Atrovate AI & Computer Vision Hackathon 2018

I’m grateful and excited that our team is one of the three winners of the Atrovate Artificial Intelligence and Computer Vision Hackathon 2018.

We developed a system that performs real-time detection of violence, nudity, drugs and gambling occurring in a video stream.

In the screenshot, me playing around with our app: it uses the laptop webcam as input video stream and blurs it when something bad is detected. 🙂

Thank you Intel Movidius for the prize.