| title | Weather Whiplash API |
|---|---|
| emoji | 🏎️ |
| colorFrom | red |
| colorTo | gray |
| sdk | docker |
| app_port | 7860 |
Track-surface intelligence for tyre decisions. Upload racing footage to classify the surface, smooth it into a Track Wetness Index, detect whether conditions are changing, and estimate when to pit.
- Classifies each frame as
dry,damp,wet, orstanding_waterwithopenai/clip-vit-base-patch32. - Derives
DRYING,WETTING, orSTABLEfrom the smoothed signal—not from a single image. - Shows frame quality, uncertainty, weather fusion, crossover timing, and a hysteresis based tyre recommendation.
FastAPI, PyTorch/Transformers, OpenCV, SQLite, and Next.js. The frontend and backend remain separate processes connected over HTTP.
Requires Python 3.11+ and Node.js.
python3 -m venv .venv
.venv/bin/pip install -r backend/requirements.txt
cd frontend && npm ci && cd ..
# Terminal 1
cd backend && ../.venv/bin/python -m uvicorn app.main:app --port 8000
# Terminal 2
cd frontend && npm run devOpen http://localhost:3000. The first backend start downloads and warms the model.
Configuration and deployment
The application uses no runtime API key. Keep optional Hugging Face write tokens in your ignored .env file or in your shell environment; use the committed .env.example only as a template.
Future Prospects
Weather Whiplash is intended to evolve from a zero-shot vision baseline into a Formula 1-specific track-condition and tyre-strategy intelligence system.
Future development will focus on:
1.Hugging Face model integration — evaluate and integrate stronger vision models from the Hugging Face ecosystem and move beyond the current zero-shot CLIP approach.
2.F1-specific dataset and fine-tuning — build a human-verified dataset from Formula 1 racing footage and track imagery covering dry, damp, wet, and standing_water, then fine-tune suitable Hugging Face vision models using PyTorch and Transformers.
3.Track-condition intelligence — improve detection of surface wetness, standing water, spray, drying lines, and changing grip conditions across consecutive frames. Temporal modelling — combine frame-level predictions with temporal models and sequence analysis to produce more reliable wetting/drying trends.
4.Tyre-strategy prediction — combine track condition, wetness trend, weather, lap-time behaviour, and uncertainty to recommend when slicks, intermediates, or full wets are likely to become optimal.
5.F1 telemetry integration — incorporate telemetry and race data such as lap times, sector performance, tyre compounds, pit stops, race position, and track conditions to improve strategic predictions.
6.Weather-data fusion — combine computer-vision predictions with rainfall, temperature, humidity, wind, and forecast data to create a more complete track-condition model.
7.Circuit-specific intelligence — incorporate circuit layouts, sectors, racing lines, surface characteristics, and historical race-weather data so that the model understands that different circuits can behave differently under the same weather conditions.
8.Multimodal race analysis — combine video, telemetry, weather, historical race data, and circuit information into a unified race-engineering report. Real-time analysis — extend the current uploaded-clip workflow toward continuous live analysis of racing footage.
9.Uncertainty-aware predictions — preserve the system's ability to refuse unreliable predictions and explicitly report when the available evidence is insufficient.
The long-term objective is to combine:
Computer Vision + Hugging Face Models + Temporal Analysis + Weather Data + F1 Telemetry + Circuit Intelligence + Tyre Strategy into a single system capable of analysing racing footage and producing an engineering-style track-condition and tyre-strategy report.
Licence
MIT. Dataset sources and attribution are documented in docs/DATASET.md. The compiled dataset is available on Hugging Face at mrcreoid/weather-whiplash-surfaces