Annotation-Efficient Object Detection: Unsupervised Discovery to Active Learning

Explore annotation-efficient approaches to object detection in this comprehensive seminar by AI Research Scientist Huy V. Vo from FAIR Labs, Meta. Delve into alternatives to fully-supervised object detection that require less or no manual annotation, including unsupervised object discovery and active learning strategies. Learn about optimization-based approaches, ranking methods, and seed-growing techniques that leverage self-supervised transformers for identifying and localizing similar objects in image collections without manual annotation. Discover the BiB active learning strategy, designed to combine weakly-supervised and active learning for training object detectors, offering an improved balance between annotation cost and effectiveness compared to traditional methods. Gain insights into cutting-edge research on learning problems with limited supervision, essential for developing intelligent systems like autonomous vehicles and robots.