Abstract
This document describes a distributed computing platform for automatically performing feature extraction, signal extraction, and/or metadata extraction from large collections of media files, including videos, audio recordings, and multimedia content. A cloud-based processing system is described that uses a “Bring-Your-Own-Signal” framework to enable developers to submit custom extraction programs, algorithms, machine learning models, or processing graphs directly into the central management system. The system automatically validates and sets up submitted extraction algorithms for use. The system may support resource sharing and improve scalability by dispatching extraction tasks to execute on client-provided hardware, and receiving, from the client-provided hardware, only the resulting extracted signal data and omitting the large, raw media files so as to conserve network bandwidth. The system may, for example, continuously and automatically evaluate new media files upon ingestion into a storage repository or upon converting such media files into additional formats. The decentralized, scalable setup described herein accelerates metadata generation for video analytics datasets and attributes that may be used for training generative artificial intelligence (AI) systems, or other AI systems.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Verlani, Pooja; Adsumilli, Balu; and Kanetkar, Shruti, "SCALABLE PER-SCENE MEDIA SIGNAL GENERATION USING CONFIGURABLE SIGNAL EXTRACTION ALGORITHMS", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/10296