Microsoft’s BEiT-3 Foundation Model: A ‘Big Convergence of Language, Vision, and Multimodal Pretraining’ That Achieves SOTA Results on Popular Benchmarks

In the new paper Image as a Foreign Language: BEiT Pretraining for All Vision and Vision-Language Tasks, a Microsoft research team presents BEiT-3, a general-purpose state-of-the-art multimodal foundation model for both vision and vision-language tasks that advances the big convergence of backbone architectures, pretraining tasks, and model scaling.

ai artificial intelligence benchmarks big computer vision & graphics convergence deep-neural-networks foundation model language machine learning machine learning & data science microsoft ml multimodal multimodal learning popular research sota technology vision