CVPR Daily - Monday

Liliane’s picks of the day (sessions 1, 2, 3): Liliane Momeni is currently a second year PhD student in the VGG group at the University of Oxford, supervised by Andrew Zisserman. Her research focuses on understanding vision and language, particularly in the context of sign language. What do I work on? “ I work in a small team with Gül Varol , Samuel Albanie and Triantafyllos Afouras on sign language understanding. Sign languages are visual- spatial languages that have evolved among deaf communities and differ considerably from spoken languages. ” (1) Discovering Relationships between Object Categories via Universal Canonical Maps (1) Improving Sign Language Translation With Monolingual Data by Sign Back-Translation (1) Temporal-Relational CrossTransformers for Few-Shot Action Recognition (2) Re-Labelling ImageNet: From Single to Multi-Labels, From Global to Localised labels (3) Instance Localization for Self-supervised Detection Pretraining (3) A Large-Scale Study on Unsupervised Spatiotemporal Representation Learning (3) On Semantic Similarity in Video Retrieval For today, Monday 21 2 Liliane’s Picks DAILY CVPR Monday Why do I think it’s important? “Despite hundreds of sign languages and millions of people using them, almost all communication platforms are still designed to exclusively support spoken or written languages . I believe it is vital to try to adapt these platforms to empower all users. Our work so far takes steps towards enabling applications such as helping virtual assistants respond to signed “wake words” and providing efficient search of sign language videos. ” Why is it an interesting technical problem? “Research in sign language addresses several key problems in machine learning that I hope to contribute to, including hierarchical sequence modeling, multimodal representation learning, and learning from weak supervision . Despite the recent progress in fine-grained gesture classification and machine translation, automatic sign language understanding still remains far from solved and a great deal of exciting work lies ahead. ” Learn more about Liliane’s work! Join her poster and Q&A session presenting “Read and Attend: Temporal Localisation in Sign Language Videos” on Friday from 11:00 - 13:30 BST.