Synthetic-to-real Language-guided Action/Activity recognition
Monday 15 July 2024

Synthetic datasets are a potentially unlimited source of new data. On the other hand, the domain gap between synthetic and real data is relevant, and often a model trained on the first performs poorly on the latter. This project aims to learn temporal dynamics, and disentangling synthetic appearance using large language models.

Artificial Intelligence
Machine learning
Natural language processing
Computer vision
Neural networks
Activity recognition and understanding
Scene understanding
Deep learning