Grounded Multi-Style Video Captioning Agent
Grounded Multi-Style Video Captioning Agent is a Dockerized AI system for AMD Developer Hackathon ACT II, Track 2. It reads video-captioning tasks from `/input/tasks.json`, downloads each video, samples resized frames across the full clip using ffmpeg, and uses a Fireworks-hosted vision model to understand the scene. The agent first creates a grounded factual summary of visible settings, subjects, actions, objects, mood, and notable details. It then converts that summary into four required caption styles: formal, sarcastic, humorous-tech, and humorous-non-tech. To improve reliability, the system includes JSON repair, raw model-response handling, and deterministic grounded fallbacks, so even imperfect model outputs can still produce valid `/output/results.json` captions without generic placeholders.
Category tags:
For Innovators & Creators