At Kino AI (https://trykino.com), we are seeking a full or part-time machine learning engineer for our initial product offering, which seeks to automatically index and organize raw video footage for professional filmmakers via metadata generation.
Everyone who works in video faces the footage management problem. Hollywood film editors, YouTube content creators, and indie documentarians all trudge through endless amounts of footage to find the best moments, retrieve particular scenes, and give shape to their story. At Kino AI, we're building tools and workflows that simplify this process via intelligent search and automatic video labeling. See our website https://trykino.com) for more info.
Our smart tools are powered by our AI Engine. Given raw user video, the Engine outputs structured metadata. This metadata includes scene transcriptions, visual descriptions, periods of inactivity, face tracking, and more. As a Kino Machine Learning Engineer, your job is to track new forms of useful metadata, improve engine performance, and work with the rest of the Kino team to build an excellent product.
Task Overview:
Given a large amount of raw HD video footage, generate useful and well-defined metadata for each video file, each distinct moment in each video file, and/or each frame in each video file.
Your task is to create a well-tested and performant AI metadata engine. You are encouraged to try many different approaches as you tackle the various types of metadata that are worth capturing. The engine should have robust I/O (many video filetypes supported, etc), be easy to monitor (report job progress via tqdm (https://github.com/tqdm/tqdm) or similar), and should output metadata in an easy-to-understand format (JSON, YAML, etc).
View the full challenge at this Notion site: https://www.notion.so/trykino/Kino-AI-Challenge-Metadata-Engine-Hackathon-bbf5b2ec6c904a86a4ff3ecddd57e320?pvs=4