How to Build a Huge Dataset Without Spending a Fortune
summary
Describes how the HaGRID project collected and annotated a very large gesture dataset for machine learning. Covers practical methods and processes used to gather, label and publish the dataset while minimizing costs.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 3:19Crowdsourcing platforms
- 11:18Image collection conditions
- 20:34Labeling quality issues
- 29:33Automating label aggregation
- 34:35Dataset publication expansion
- 44:15Quality assessment processes
- 50:20Q&A
related talks
34mThe Recipe for Perfect Annotation in Computer Vision
10mA large, crowdsourced evaluation of gesture generation systems on common data: The GENEA ...
37mBest Practices for Building AI Datasets
05mCollecting Massive Data via Crowdsourcing
28mData Curation: Transforming Bytes into AI Gold
34mDeep Learning at Scale
54mHow to handle the luxury of having too much training data
09mGenerating Synthetic Data with AI