As a few other other have pointed out, the data from this game is open-source and accessible in number of ways [1].
For an introductory exercise to deep learning for image classification, it's a great alternative (or follow-up) to the classic MNIST dataset [2], which serves as a common "Hello, world" for image-based ML.
We trained and deployed such a model for a demo of our mobile ML tool [3][4]. Feel free to ping me if you're interested, would love to chat.
I'm excited to share this project with you all today. It's still in beta, but with positive notes from early users I wanted to open it up to the community.
Pallet is a mobile-first machine learning platform that enables you to instantly turn computer vision models into shareable apps, and access them anytime from a single interface.
As an application developer, I was drawn to deep learning for computer vision for the seemingly magical feature of giving everyday apps the ability to "see". However, while there are a number of resources that can teach you how to build & train deep learning models for object recognition tasks, I've found far fewer resources that facilitate deploying those same models as real-world apps. There are even a handful of amazing "no-code" applications for developing image classification models, like Lobe.ai [1] or Google Cloud's AutoML Vision [2], but no comparable applications for deployment.
Moreover, given how fast the ML world has been moving the last few years, it can be a challenge to not only keep up with state of the art models, but understand how to use them in practice.
In an effort to make this tech a bit more accessible, I started building Pallet to automate hosting, serving, integrating, and updating models targeted for mobile devices.
With Pallet you can:
• Deploy custom machine learning models to mobile without code, and try them in the real world,
• Share a link to any model with one tap,
• Make predictions with state-of-the-art models anytime, and
• Explore a number of models made by the ML community
It's a first step towards simplifying the process of building, deploying, and sharing custom AI-enabled apps, particularly for individual developers, machine learning engineers, students, and data scientists.
Right now the platform supports pretty much any TensorFlow image classification model with a standard signature (including those you can export from the aforementioned platforms), and you can find the Android app in the Play Store [3]. With more time and resources, I plan to improve framework and platform support.
I'd love to hear any and all feedback in the comments, or ping me on Twitter @PalletML. Thanks!
This is incredible, and an entertaining use case for stylegan. Big fan of the first example track (Kupla x j'san - raindrops) which caught my attention. Keep up the great work!
Hacker News Daily is a daily collection of all of the stories that have reached the homepage of Hacker News since the morning (currently 10:00am ET or 15:00 UTC).
