Text-to-image generation aims at generating realistic images which are semantically consistent with the given text. Previous works mainly adopt the multi-stage architecture by stacking generator-discriminator pairs to engage multiple adversarial training, where the text semantics used to provide generation guidance remain static across all stages. This work argues that text features at each stage should be adaptively re-composed conditioned on the status of the historical stage (i.e., historical stage's text and image features) to provide diversified and accurate semantic guidance during the coarse-to-fine generation process. We thereby propose a novel Dynamical Semantic Evolution GAN (DSE-GAN) to re-compose each stage's text features under a novel single adversarial multi-stage architecture. Specifically, we design (1) Dynamic Semantic Evolution (DSE) module, which first aggregates historical image features to summarize the generative feedback, and then dynamically selects words required to be re-composed at each stage as well as re-composed them by dynamically enhancing or suppressing different granularity subspace's semantics. (2) Single Adversarial Multi-stage Architecture (SAMA), which extends the previous structure by eliminating complicated multiple adversarial training requirements and therefore allows more stages of text-image interactions, and finally facilitates the DSE module. We conduct comprehensive experiments and show that DSE-GAN achieves 7.48\% and 37.8\% relative FID improvement on two widely used benchmarks, i.e., CUB-200 and MSCOCO, respectively.
Knowledge Graphs (KGs) are becoming increasingly essential infrastructures in many applications while suffering from incompleteness issues. The KG completion task (KGC) automatically predicts missing facts based on an incomplete KG. However, existing methods perform unsatisfactorily in real-world scenarios. On the one hand, their performance will dramatically degrade along with the increasing sparsity of KGs. On the other hand, the inference procedure for prediction is an untrustworthy black box. This paper proposes a novel explainable model for sparse KGC, compositing high-order reasoning into a graph convolutional network, namely HoGRN. It can not only improve the generalization ability to mitigate the information insufficiency issue but also provide interpretability while maintaining the model's effectiveness and efficiency. There are two main components that are seamlessly integrated for joint optimization. First, the high-order reasoning component learns high-quality relation representations by capturing endogenous correlation among relations. This can reflect logical rules to justify a broader of missing facts. Second, the entity updating component leverages a weight-free Graph Convolutional Network (GCN) to efficiently model KG structures with interpretability. Unlike conventional methods, we conduct entity aggregation and design composition-based attention in the relational space without additional parameters. The lightweight design makes HoGRN better suitable for sparse settings. For evaluation, we have conducted extensive experiments-the results of HoGRN on several sparse KGs present impressive improvements (9% MRR gain on average). Further ablation and case studies demonstrate the effectiveness of the main components. Our codes will be released upon acceptance.
In recommender system, some feature directly affects whether an interaction would happen, making the happened interactions not necessarily indicate user preference. For instance, short videos are objectively easier to be finished even though the user does not like the video. We term such feature as confounding feature, and video length is a confounding feature in video recommendation. If we fit a model on such interaction data, just as done by most data-driven recommender systems, the model will be biased to recommend short videos more, and deviate from user actual requirement. This work formulates and addresses the problem from the causal perspective. Assuming there are some factors affecting both the confounding feature and other item features, e.g., the video creator, we find the confounding feature opens a backdoor path behind user item matching and introduces spurious correlation. To remove the effect of backdoor path, we propose a framework named Deconfounding Causal Recommendation (DCR), which performs intervened inference with do-calculus. Nevertheless, evaluating do calculus requires to sum over the prediction on all possible values of confounding feature, significantly increasing the time cost. To address the efficiency challenge, we further propose a mixture-of experts (MoE) model architecture, modeling each value of confounding feature with a separate expert module. Through this way, we retain the model expressiveness with few additional costs. We demonstrate DCR on the backbone model of neural factorization machine (NFM), showing that DCR leads to more accurate prediction of user preference with small inference time cost.
Rumor detection has become an emerging and active research field in recent years. At the core is to model the rumor characteristics inherent in rich information, such as propagation patterns in social network and semantic patterns in post content, and differentiate them from the truth. However, existing works on rumor detection fall short in modeling heterogeneous information, either using one single information source only (e.g. social network, or post content) or ignoring the relations among multiple sources (e.g. fusing social and content features via simple concatenation). Therefore, they possibly have drawbacks in comprehensively understanding the rumors, and detecting them accurately. In this work, we explore contrastive self-supervised learning on heterogeneous information sources, so as to reveal their relations and characterize rumors better. Technically, we supplement the main supervised task of detection with an auxiliary self-supervised task, which enriches post representations via post self-discrimination. Specifically, given two heterogeneous views of a post (i.e. representations encoding social patterns and semantic patterns), the discrimination is done by maximizing the mutual information between different views of the same post compared to that of other posts. We devise cluster-wise and instance-wise approaches to generate the views and conduct the discrimination, considering different relations of information sources. We term this framework as Self-supervised Rumor Detection (SRD). Extensive experiments on three real-world datasets validate the effectiveness of SRD for automatic rumor detection on social media.
Action recognition from videos, i.e., classifying a video into one of the pre-defined action types, has been a popular topic in the communities of artificial intelligence, multimedia, and signal processing. However, existing methods usually consider an input video as a whole and learn models, e.g., Convolutional Neural Networks (CNNs), with coarse video-level class labels. These methods can only output an action class for the video, but cannot provide fine-grained and explainable cues to answer why the video shows a specific action. Therefore, researchers start to focus on a new task, Part-level Action Parsing (PAP), which aims to not only predict the video-level action but also recognize the frame-level fine-grained actions or interactions of body parts for each person in the video. To this end, we propose a coarse-to-fine framework for this challenging task. In particular, our framework first predicts the video-level class of the input video, then localizes the body parts and predicts the part-level action. Moreover, to balance the accuracy and computation in part-level action parsing, we propose to recognize the part-level actions by segment-level features. Furthermore, to overcome the ambiguity of body parts, we propose a pose-guided positional embedding method to accurately localize body parts. Through comprehensive experiments on a large-scale dataset, i.e., Kinetics-TPS, our framework achieves state-of-the-art performance and outperforms existing methods over a 31.10% ROC score.
In this paper, we abandon the dominant complex language model and rethink the linguistic learning process in the scene text recognition. Different from previous methods considering the visual and linguistic information in two separate structures, we propose a Visual Language Modeling Network (VisionLAN), which views the visual and linguistic information as a union by directly enduing the vision model with language capability. Specially, we introduce the text recognition of character-wise occluded feature maps in the training stage. Such operation guides the vision model to use not only the visual texture of characters, but also the linguistic information in visual context for recognition when the visual cues are confused (e.g. occlusion, noise, etc.). As the linguistic information is acquired along with visual features without the need of extra language model, VisionLAN significantly improves the speed by 39% and adaptively considers the linguistic information to enhance the visual features for accurate recognition. Furthermore, an Occlusion Scene Text (OST) dataset is proposed to evaluate the performance on the case of missing character-wise visual cues. The state of-the-art results on several benchmarks prove our effectiveness. Code and dataset are available at https://github.com/wangyuxin87/VisionLAN.
Real-world recommender system needs to be regularly retrained to keep with the new data. In this work, we consider how to efficiently retrain graph convolution network (GCN) based recommender models, which are state-of-the-art techniques for collaborative recommendation. To pursue high efficiency, we set the target as using only new data for model updating, meanwhile not sacrificing the recommendation accuracy compared with full model retraining. This is non-trivial to achieve, since the interaction data participates in both the graph structure for model construction and the loss function for model learning, whereas the old graph structure is not allowed to use in model updating. Towards the goal, we propose a \textit{Causal Incremental Graph Convolution} approach, which consists of two new operators named \textit{Incremental Graph Convolution} (IGC) and \textit{Colliding Effect Distillation} (CED) to estimate the output of full graph convolution. In particular, we devise simple and effective modules for IGC to ingeniously combine the old representations and the incremental graph and effectively fuse the long-term and short-term preference signals. CED aims to avoid the out-of-date issue of inactive nodes that are not in the incremental graph, which connects the new data with inactive nodes through causal inference. In particular, CED estimates the causal effect of new data on the representation of inactive nodes through the control of their collider. Extensive experiments on three real-world datasets demonstrate both accuracy gains and significant speed-ups over the existing retraining mechanism.
Existing scene text removal methods mainly train an elaborate network with paired images to realize the function of text localization and background reconstruction simultaneously, but there exists two problems: 1) lacking the exhaustive erasure of text region and 2) causing the excessive erasure to text-free areas. To handle these issues, this paper provides a novel ProgrEssively Region-based scene Text eraser (PERT), which introduces region-based modification strategy to progressively erase the pixels in only text region. Firstly, PERT decomposes the STR task to several erasing stages. As each stage aims to take a further step toward the text-removed image rather than directly regress to the final result, the decomposed operation reduces the learning difficulty in each stage, and an exhaustive erasure result can be obtained by iterating over lightweight erasing blocks with shared parameters. Then, PERT introduces a region-based modification strategy to ensure the integrity of text-free areas by decoupling text localization from erasure process to guide the removal. Benefiting from the simplicity architecture, PERT is a simple and strong baseline, and is easy to be followed and developed. Extensive experiments demonstrate that PERT obtains the state-of-the-art results on both synthetic and real-world datasets. Code is available athttps://github.com/wangyuxin87/PERT.
Cross-modal correlation provides an inherent supervision for video unsupervised representation learning. Existing methods focus on distinguishing different video clips by visual and audio representations. We human visual perception could attend to regions where sounds are made, and our auditory perception could also ground their frequencies of sounding objects, which we call bidirectional local correspondence. Such supervision is intuitive but not well explored in the contrastive learning framework. This paper introduces a pretext task, Cross-Modal Attention Consistency (CMAC), for exploring the bidirectional local correspondence property. The CMAC approach aims to align the regional attention generated purely from the visual signal with the target attention generated under the guidance of acoustic signal, and do a similar alignment for frequency grounding on the acoustic attention. Accompanied by a remoulded cross-modal contrastive loss where we consider additional within-modal interactions, the CMAC approach works effectively for enforcing the bidirectional alignment. Extensive experiments on six downstream benchmarks demonstrate that CMAC can improve the state-of-the-art performance on both visual and audio modalities.
Occluded person re-identification (Re-ID) is a challenging task as persons are frequently occluded by various obstacles or other persons, especially in the crowd scenario. To address these issues, we propose a novel end-to-end Part-Aware Transformer (PAT) for occluded person Re-ID through diverse part discovery via a transformer encoderdecoder architecture, including a pixel context based transformer encoder and a part prototype based transformer decoder. The proposed PAT model enjoys several merits. First, to the best of our knowledge, this is the first work to exploit the transformer encoder-decoder architecture for occluded person Re-ID in a unified deep model. Second, to learn part prototypes well with only identity labels, we design two effective mechanisms including part diversity and part discriminability. Consequently, we can achieve diverse part discovery for occluded person Re-ID in a weakly supervised manner. Extensive experimental results on six challenging benchmarks for three tasks (occluded, partial and holistic Re-ID) demonstrate that our proposed PAT performs favorably against stat-of-the-art methods.