XGPT: Cross-modal Generative Pre-Training for Image Captioning

Abstract

While many BERT-based cross-modal pre-trained models produce excellent results on downstream understanding tasks like image-text retrieval and VQA, they cannot be applied to generation tasks directly. In this paper, we propose XGPT, a new method of Cross-modal Generative Pre-Training for Image Captioning that is designed to pre-train text-to-image caption generators through three novel generation tasks, including Image-conditioned Masked Language Modeling (IMLM), Image-conditioned Denoising Autoencoding (IDA), and Text-conditioned Image Feature Generation (TIFG). As a result, the pre-trained XGPT can be fine-tuned without any task-specific architecture modifications to create state-of-the-art models for image captioning. Experiments show that XGPT obtains new state-of-the-art results on the benchmark datasets, including COCO Captions and Flickr30k Captions. We also use XGPT to generate new image captions as data augmentation for the image retrieval task and achieve significant improvement on all recall metrics.

12 pages, 3 figures, 7 tables

References in corpus (9)

• XLNet: Generalized Autoregressive Pretraining for Language Understanding
• Microsoft COCO Captions: Data Collection and Evaluation Server
• VisualBERT: A Simple and Performant Baseline for Vision and Language
• Unified Language Model Pre-training for Natural Language Understanding and Generation
• VL-BERT: Pre-training of Generic Visual-Linguistic Representations
• MASS: Masked Sequence to Sequence Pre-training for Language Generation
• UNITER: UNiversal Image-TExt Representation Learning
• Unicoder-VL: A Universal Encoder for Vision and Language by Cross-modal Pre-training
• Unicoder: A Universal Language Encoder by Pre-training with Multiple Cross-lingual Tasks

Cited by in corpus (10)

• Large-Scale Adversarial Training for Vision-and-Language Representation Learning
• Unifying Vision-and-Language Tasks via Text Generation
• WenLan: Bridging Vision and Language by Large-Scale Multi-Modal Pre-Training
• HERO: Hierarchical Encoder for Video+Language Omni-representation Pre-training
• M3P: Learning Universal Representations via Multitask Multilingual Multimodal Pre-training
• RepsNet: Combining Vision with Language for Automated Medical Reports
• Pretrained Language Models for Text Generation: A Survey
• Pre-Trained Models: Past, Present and Future
• Kaleido-BERT: Vision-Language Pre-training on Fashion Domain
• Modeling Text-visual Mutual Dependency for Multi-modal Dialog Generation