Self-Training Vision Language BERTs with a Unified Conditional Model
This addresses the data bottleneck for vision-language pretraining, enabling more scalable training with less labeled data, though it is incremental as it builds on existing self-training and generation methods.
The paper tackles the problem of training vision-language BERTs without paired data by proposing a self-training approach using a unified conditional model for zero-shot generation, achieving competitive or better performance with only 300k unlabeled extra data compared to models trained with 3 million extra data.
Natural language BERTs are trained with language corpus in a self-supervised manner. Unlike natural language BERTs, vision language BERTs need paired data to train, which restricts the scale of VL-BERT pretraining. We propose a self-training approach that allows training VL-BERTs from unlabeled image data. The proposed method starts with our unified conditional model -- a vision language BERT model that can perform zero-shot conditional generation. Given different conditions, the unified conditional model can generate captions, dense captions, and even questions. We use the labeled image data to train a teacher model and use the trained model to generate pseudo captions on unlabeled image data. We then combine the labeled data and pseudo labeled data to train a student model. The process is iterated by putting the student model as a new teacher. By using the proposed self-training approach and only 300k unlabeled extra data, we are able to get competitive or even better performances compared to the models of similar model size trained with 3 million extra image data.