[Adapt] [Seminar] Spatial Commonsense Knowledge in Vision Models

[罗嘉鸣]

Hi Adapters,

Spatial commonsense knowledge is referred to spatial position and relationship between objects, such as the relative size of an ant and an elephant or the relative position of a man and a car. Although pre-trained language models achieve great success in many NLP tasks, they seem to be ineffective in spatial commonsense reasoning.

On the contrary, vision models such as vision-language models and image synthesis models are more likely to contain more spatial commonsense knowledge since images naturally describe spatial relationships. In this talk, I will introduce one paper "Things not Written in Text: Exploring Spatial Commonsense from Visual Signals", which is accepted by ACL 2022.

Wish you would like it!

Time: Wed 4:00 pm

Venue: SEIEE 3-404

Best Regards,

Leo