XGPT: Cross-modal Generative Pre-Training for Image Captioning
This addresses the need for effective cross-modal generation in image captioning, offering a novel pre-training approach with broad applicability.
The paper tackled the problem of applying cross-modal pre-trained models to generation tasks by proposing XGPT, a method for pre-training text-to-image caption generators, which achieved state-of-the-art results on COCO Captions and Flickr30k Captions datasets and improved image retrieval through data augmentation.
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.