E2E-VLP: End-to-end Vision-language Pre-training Enhanced By Visual Learning

Xu Haiyang, Yan Ming, Li Chenliang, Bi Bin, Huang Songfang, Xiao Wenming, Huang Fei. Arxiv 2021

[Paper]    
Efficiency And Optimization Model Architecture Multimodal Models Pretraining Methods Tools Training Techniques Transformer

Vision-language pre-training (VLP) on large-scale image-text pairs has achieved huge success for the cross-modal downstream tasks. The most existing pre-training methods mainly adopt a two-step training procedure, which firstly employs a pre-trained object detector to extract region-based visual features, then concatenates the image representation and text embedding as the input of Transformer to train. However, these methods face problems of using task-specific visual representation of the specific object detector for generic cross-modal understanding, and the computation inefficiency of two-stage pipeline. In this paper, we propose the first end-to-end vision-language pre-trained model for both V+L understanding and generation, namely E2E-VLP, where we build a unified Transformer framework to jointly learn visual representation, and semantic alignments between image and text. We incorporate the tasks of object detection and image captioning into pre-training with a unified Transformer encoder-decoder architecture for enhancing visual learning. An extensive set of experiments have been conducted on well-established vision-language downstream tasks to demonstrate the effectiveness of this novel VLP paradigm.

Similar Work