README.md

---
title: Demo 2025
emoji: 📊
colorFrom: green
colorTo: red
sdk: docker
pinned: false
license: mit
---

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

## Mission-guided detections

1. Call `POST /process_video` with fields `video` (file), `prompt` (mission text), and optional `detector` (`owlv2` or `hf_yolov8`). The response is an MP4 stream containing the annotated frames.
2. Call `POST /mission_summary` with the same fields to receive JSON containing the structured mission plan plus the natural-language summary. This second endpoint isolates the OpenAI call, keeping the video response clean.
3. Under the hood the mission text still feeds into the OpenAI (`gpt-4o-mini`) reasoning step that ranks the YOLO/COCO classes. Place your API key inside `.env` as either `OPENAI_API_KEY=...` or `OpenAI-API: ...`; the server loads it automatically on startup.
4. The top scored classes drive OWLv2 or YOLOv8 to align detections with the mission, and the detection log is summarized via another OpenAI call when requested.