Image as a Foreign Language: BEiT Pretraining for All Vision and Vision-Language Tasks
This work addresses the need for unified pretraining models in AI by providing a versatile solution for researchers and practitioners in computer vision and multimodal learning, though it builds incrementally on existing convergence trends.
The authors tackled the problem of creating a general-purpose multimodal foundation model by introducing BEiT-3, which achieves state-of-the-art performance on various vision and vision-language tasks, such as object detection on COCO, semantic segmentation on ADE20K, and image classification on ImageNet.
A big convergence of language, vision, and multimodal pretraining is emerging. In this work, we introduce a general-purpose multimodal foundation model BEiT-3, which achieves state-of-the-art transfer performance on both vision and vision-language tasks. Specifically, we advance the big convergence from three aspects: backbone architecture, pretraining task, and model scaling up. We introduce Multiway Transformers for general-purpose modeling, where the modular architecture enables both deep fusion and modality-specific encoding. Based on the shared backbone, we perform masked "language" modeling on images (Imglish), texts (English), and image-text pairs ("parallel sentences") in a unified manner. Experimental results show that BEiT-3 obtains state-of-the-art performance on object detection (COCO), semantic segmentation (ADE20K), image classification (ImageNet), visual reasoning (NLVR2), visual question answering (VQAv2), image captioning (COCO), and cross-modal retrieval (Flickr30K, COCO).