Conditioned diffusion models have demonstrated state-of-the-art text-to-image synthesis capacity. Recently, most works focus on synthesizing independent images; While for real-world applications, it is common and necessary to generate a series of coherent images for story-stelling. In this work, we mainly focus on story visualization and continuation tasks and propose AR-LDM, a latent diffusion model auto-regressively conditioned on history captions and generated images. Moreover, AR-LDM can generalize to new characters through adaptation. To our best knowledge, this is the first work successfully leveraging diffusion models for coherent visual story synthesizing. It also extends the text-conditioned method to multimodal conditioning. Quantitative results show that AR-LDM achieves SoTA FID scores on PororoSV, FlintstonesSV, and the adopted challenging dataset VIST containing natural images. Large-scale human evaluations show that AR-LDM has superior performance in terms of quality, relevance, and consistency. Code available at this https URL

Deep neural network has adequately revealed its superiority of solving various tasks in Natural Language Processing, especially for relation classification. However, unlike traditional feature-engineering methods that targetedly extract well-designed features for specific task, the diversity of input format for deep learning is limited; word sequence as input is the frequently used setting. Therefore, the input of neural network, to some extent, lacks pertinence. For relation classification task, it is not uncommon that, without specific entity pair, a sentence contains various relation types; therefore, entity pair indicates the distribution of the crucial information in input sentence for recognizing specific relation. Aiming at this characteristic, in this paper, several strategies are proposed to integrate entity pair information into the application of deep learning in relation classification task, in a way to provide definitive learning direction for neural network. Experimental results on the SemEval-2010 Task 8 dataset show that our method outperforms most of the state-of-the-art models, without external linguistic features.