Announcing tools for the Infinity API
We’re so excited to announce our latest Infinity API release! Not only is the rendering 2x faster with double the number of exercises in the library, but most notably… we’ve publicly released tools to make it easier than ever to work with the API! We take a closer look at the repos and code tutorials below.
Sign up for the API
Sign up for the API here. Or contact me if you have any questions or want to chat about your needs: andrew@toinfinity.ai. I read every email!
Infinity Tools
This is a small Python library that wraps the Infinity REST API with various convenience tools and functions. To install, create a new virtual environment and follow the instructions here. You’re now setup to run the Jupyter notebooks below!
Infinity Tutorials
These demo notebooks illustrate several common workflows with the VisionFit (image/video) and SenseFit (Apple Watch / IMU) APIs.
VisionFit README
+ Explore API Parameters Demo Notebook
+ Workflow Demo Notebook
+ Large Job Demo Notebook
+ Large Sweep Job Demo Notebook
+ Rep Counting Demo Notebook
SenseFit README
+ Explore API Parameters Demo Notebook
+ Large Job Demo Notebook
+ Rep Counting Demo Notebook
3 Highlights
1 | Generate a dataset according to your own specifications
One of the magical things about a synthetic dataset is that it can be designed according to precise specifications. As an illustrative example, the VisionFit Large Job Demo Notebook shows how users can generate a dataset of 1000 videos with a normal distribution of camera heights (centered at 1.2 meters), a normal distribution of lighting conditions (centered at 300 watts), and uniform distributions for other parameters like rep speed. Users of the Infinity API can change these distributions as appropriate for their own applications.
2 | Bootstrapping ML models
Another reason to love synthetic data is that it can be used to bootstrap models where real-world data is expensive, time-consuming, or otherwise difficult to collect. Using both VisionFit (videos) and SenseFit (Apple Watch) APIs, we show how to train an RNN rep-counting model purely on synthetic data and successfully deploy it on real-world data. Give it a try! We’ve made it easy to use your own real-world videos or Apple Watch datasets for inference.
3 | Fix Failure Cases
Last but not least, synthetic data can be used to quickly and iteratively improve a model’s performance. An illustrative example of this workflow is provided in the SenseFit Rep Counting Demo Notebook. In it, we improve the performance of a baseline rep counting model by generating a new synthetic dataset specifically designed to match the model’s failure cases. Re-training the model with this new synthetic data “solves” the failure cases and results in a model with improved real-world performance.
Infinity AI. For ML engineers, by ML engineers.
We’re a small team of engineers who build tools to accelerate progress in ML. We’d love to share our journey of building this company with you. Follow us on LinkedIn or get in touch (andrew@toinfinity.ai).
Website: https://toinfinity.ai/
Email list: sign up here
Our Past Releases:
January — InfiniteRep: An open-source synthetic dataset for remote fitness and PT applications
February — Announcing the Infinity API: Data at your fingertips
March — New Infinity API Features
