Facebook has invested a lot of resources in AI artificial intelligence, and the results have been open source. Recently, its deep learning framework Pythia has been open sourced for image and language analysis, which is convenient for establishing, copying and testing related artificial intelligence models.
According to Facebook, Pythia is a tool based on the PyTorch machine learning framework for visual and linguistic analysis, such as processing visually impaired data related queries, or automatically generating image introductions. This tool also integrates various award-winning technologies from Facebook AI, such as LoRRA visual and language models. Pythia also supports decentralized training and multiple data sets with built-in VQA, VizWiz, TextVQA and VisualDialog footage. After open source, Facebook said it will introduce more tools, data sets and other reference models in the future.
According to Facebook, this will make it easier for the development community to build and test benchmarks on successful systems. They want to remove barriers and enable researchers to develop new ways of communicating people and smart machines faster. This work should also help researchers develop adaptive artificial intelligence that integrates multiple understandings into more context-based multimodal understanding.