WebSSL

April 25, 2011 - Technology media marktechpost published a blog post yesterday (April 24) reporting that Meta released a series of WebSSL models with parameter sizes ranging from 300 million to 7 billion, trained on pure image data, with the aim of exploring the potential of unsupervised visual self-supervised learning (SSL) without language supervision. Represented by OpenAI's CLIP, contrasting language-image models have become the default choice for learning visual representations, excelling in multimodal tasks such as visual question and answer (VQA) and document understanding. However, subject to ...