Introduction

Start with a question you can test. Choose a measurement, find a reference, and define your evaluation before spending the whole event on an interface.

Tools and Technologies

• Data: PhysioNet, MIMIC-IV, Kaggle, and Hugging Face Datasets. MIMIC-IV requires a credentialing step. Start early if you plan to use it.
• Models and analysis: scikit-learn, PyTorch, OpenCV, MediaPipe, NeuroKit2, and MONAI.
• Working demos: FastAPI, Streamlit, and Gradio.

LLM assistants and Copilot are fine. Add a disclosure line in your README naming the tools and what they did.

Inspiration

• Gulshan et al., JAMA 2016, on diabetic retinopathy screening.
• Rajpurkar et al., 2017, CheXNet.
• Wang et al., 2017, on algorithmic principles of remote photoplethysmography.
• Any paper where the authors report error bars and failure cases.

What separates a good entry

A stated protocol. A held-out set. Error in real units. A limitations section that is not decoration. A video a stranger can follow without installing anything.

FAQ

Do I need a dataset?

No. A webcam and a friend are enough to start. Get consent for recordings and define how you will evaluate your result.

Do I need a GPU?

No. Classical signal processing runs on a laptop CPU.

Can I use a pre-trained model?

Yes, with attribution.

Can I use ChatGPT or Copilot?

Yes, with disclosure. See Rules.

Does it have to be heart rate?

No. Anything the camera or data can recover that addresses a health or care problem.

Can I submit something I already built?

Yes, if it runs today and the repository is public.

Is a phone camera allowed?

Yes.

How accurate does it have to be?

Accuracy is not scored alone. Honesty about it is. State your metric, measurement method, and limitations.

When are winners announced?

December 12, on the results page, at 9:00 AM ET.