Workshop
The Second Perception Test Challenge
Viorica Patraucean
Sun 29 Sep, midnight PDT
Following the successful 2023 edition, we organise the second Perception Test Challenge to benchmark multimodal perception models on the Perception Test (blog, github) - a diagnostic benchmark created by Google DeepMind to comprehensively probe the abilities of multimodal models across:
* video, audio, and text modalities
* four skill areas: Memory, Abstraction, Physics, Semantics
* four types of reasoning: Descriptive, Explanatory, Predictive, Counterfactual
* six computational tasks: multiple-choice video-QA, grounded video-QA, object tracking, point tracking, action localisation, sound localisation
Chat is not available.