Optical character recognition (OCR) methods have been applied to diverse tasks, e.g., street view text recognition and document analysis. Recently, zero-shot OCR has piqued the interest of the research community because it considers a practical OCR scenario with unbalanced data distribution. However, there is a lack of benchmarks for evaluating such zero-shot methods that apply a divide-and-conquer recognition strategy by decomposing characters into radicals. Meanwhile, radical recognition, as another important OCR task, also lacks radical-level annotation for model training. In this paper, we construct an ancient Chinese character image dataset that contains both radical-level and character-level annotations to satisfy the requirements of the above-mentioned methods, namely, ACCID, where radical-level annotations include radical categories, radical locations, and structural relations. To increase the adaptability of ACCID, we propose a splicing-based synthetic character algorithm to augment the training samples and apply an image denoising method to improve the image quality. By introducing character decomposition and recombination, we propose a baseline method for zero-shot OCR. The experimental results demonstrate the validity of ACCID and the baseline model quantitatively and qualitatively.
33D-aware face generators are commonly trained on 2D real-life face image datasets. Nevertheless, existing facial recognition methods often struggle to extract face data captured from various camera angles. Furthermore, in-the-wild images with diverse body poses introduce a high-dimensional challenge for 3D-aware generators, making it difficult to utilize data that contains complete neck and shoulder regions. Consequently, these face image datasets often contain only near-frontal face data, which poses challenges for 3D-aware face generators to construct \textit{full-head} 3D portraits. To this end, we first create the dataset {$\it{360}^{\circ}$}-\textit{Portrait}-\textit{HQ} (\textit{$\it{360}^{\circ}$PHQ}), which consists of high-quality single-view real portraits annotated with a variety of camera parameters {(the yaw angles span the entire $360^{\circ}$ range)} and body poses. We then propose \textit{3DPortraitGAN}, the first 3D-aware full-head portrait generator that learns a canonical 3D avatar distribution from the body-pose-various \textit{$\it{360}^{\circ}$PHQ} dataset with body pose self-learning. Our model can generate view-consistent portrait images from all camera angles (${360}^{\circ}$) with a full-head 3D representation. We incorporate a mesh-guided deformation field into volumetric rendering to produce deformed results to generate portrait images that conform to the body pose distribution of the dataset using our canonical generator. We integrate two pose predictors into our framework to predict more accurate body poses to address the issue of inaccurately estimated body poses in our dataset. Our experiments show that the proposed framework can generate view-consistent, realistic portrait images with complete geometry from all camera angles and accurately predict portrait body pose.
Dynamic graph data mining has gained popularity in recent years due to the rich information contained in dynamic graphs and their widespread use in the real world. Despite the advances in dynamic graph neural networks (DGNNs), the rich information and diverse downstream tasks have posed significant difficulties for the practical application of DGNNs in industrial scenarios. To this end, in this paper, we propose to address them by pre-training and present the Contrastive Pre-Training Method for Dynamic Graph Neural Networks (CPDG). CPDG tackles the challenges of pre-training for DGNNs, including generalization capability and long-short term modeling capability, through a flexible structural-temporal subgraph sampler along with structural-temporal contrastive pre-training schemes. Extensive experiments conducted on both large-scale research and industrial dynamic graph datasets show that CPDG outperforms existing methods in dynamic graph pre-training for various downstream tasks under three transfer settings.
Existing heterogeneous graph neural networks (HGNNs) have achieved great success in utilizing the rich semantic information in heterogeneous information networks (HINs). However, few works have delved into the utilization of long-range dependencies in HINs, which is extremely valuable as many real-world HINs are sparse, and each node has only a few directly connected neighbors. Although some HGNNs can utilize distant neighbors by stacking multiple layers or leveraging long meta-paths, the exponentially increased number of nodes in the receptive field or the number of meta-paths incurs high computation and memory costs. To address these issues, we investigate the importance of different meta-paths and propose Long-range Dependency based Multi-Layer Perceptron (LDMLP). Specifically, to solve the high-cost problem of leveraging long-range dependencies, LDMLP adopts a search stage to discover effective meta-paths automatically, reducing the exponentially increased number of meta-paths to a constant. To avoid the influence of specific modules on search results, LDMLP utilizes a simple architecture with only multi-layer perceptions in the search stage, improving the generalization of searched meta-paths. As a result, the searched meta-paths not only perform well in LDMLP but also enable other HGNNs like HAN and SeHGNN to perform better. Extensive experiments on eight heterogeneous datasets demonstrate that LDMLP achieves state-of-the-art performance while enjoying high efficiency and generalization, especially on sparse HINs.
Knowledge constitutes the accumulated understanding and experience that humans use to gain insight into the world. In deep learning, prior knowledge is essential for mitigating shortcomings of data-driven models, such as data dependence, generalization ability, and compliance with constraints. To enable efficient evaluation of the worth of knowledge, we present a framework inspired by interpretable machine learning. Through quantitative experiments, we assess the influence of data volume and estimation range on the worth of knowledge. Our findings elucidate the complex relationship between data and knowledge, including dependence, synergistic, and substitution effects. Our model-agnostic framework can be applied to a variety of common network architectures, providing a comprehensive understanding of the role of prior knowledge in deep learning models. It can also be used to improve the performance of informed machine learning, as well as distinguish improper prior knowledge.
Despite large-scale pre-trained language models have achieved striking results for text classificaion, recent work has raised concerns about the challenge of shortcut learning. In general, a keyword is regarded as a shortcut if it creates a superficial association with the label, resulting in a false prediction. Conversely, shortcut learning can be mitigated if the model relies on robust causal features that help produce sound predictions. To this end, many studies have explored post-hoc interpretable methods to mine shortcuts and causal features for robustness and generalization. However, most existing methods focus only on single word in a sentence and lack consideration of word-group, leading to wrong causal features. To solve this problem, we propose a new Word-Group mining approach, which captures the causal effect of any keyword combination and orders the combinations that most affect the prediction. Our approach bases on effective post-hoc analysis and beam search, which ensures the mining effect and reduces the complexity. Then, we build a counterfactual augmentation method based on the multiple word-groups, and use an adaptive voting mechanism to learn the influence of different augmentated samples on the prediction results, so as to force the model to pay attention to effective causal features. We demonstrate the effectiveness of the proposed method by several tasks on 8 affective review datasets and 4 toxic language datasets, including cross-domain text classificaion, text attack and gender fairness test.
Graph Pooling technology plays an important role in graph node classification tasks. Sorting pooling technologies maintain large-value units for pooling graphs of varying sizes. However, by analyzing the statistical characteristic of activated units after pooling, we found that a large number of units dropped by sorting pooling are negative-value units that contain useful information and can contribute considerably to the final decision. To maintain more useful information, a novel pooling technology, called Geometric Pooling (GP), was proposed to contain the unique node features with negative values by measuring the similarity of all node features. We reveal the effectiveness of GP from the entropy reduction view. The experiments were conducted on TUdatasets to show the effectiveness of GP. The results showed that the proposed GP outperforms the SOTA graph pooling technologies by 1%\sim5% with fewer parameters.
Speech-driven 3D face animation technique, extending its applications to various multimedia fields. Previous research has generated promising realistic lip movements and facial expressions from audio signals. However, traditional regression models solely driven by data face several essential problems, such as difficulties in accessing precise labels and domain gaps between different modalities, leading to unsatisfactory results lacking precision and coherence. To enhance the visual accuracy of generated lip movement while reducing the dependence on labeled data, we propose a novel framework SelfTalk, by involving self-supervision in a cross-modals network system to learn 3D talking faces. The framework constructs a network system consisting of three modules: facial animator, speech recognizer, and lip-reading interpreter. The core of SelfTalk is a commutative training diagram that facilitates compatible features exchange among audio, text, and lip shape, enabling our models to learn the intricate connection between these factors. The proposed framework leverages the knowledge learned from the lip-reading interpreter to generate more plausible lip shapes. Extensive experiments and user studies demonstrate that our proposed approach achieves state-of-the-art performance both qualitatively and quantitatively. We recommend watching the supplementary video.
Phonetic convergence describes the automatic and unconscious speech adaptation of two interlocutors in a conversation. This paper proposes a Siamese recurrent neural network (RNN) architecture to measure the convergence of the holistic spectral characteristics of speech sounds in an L2-L2 interaction. We extend an alternating reading task (the ART) dataset by adding 20 native Slovak L2 English speakers. We train and test the Siamese RNN model to measure phonetic convergence of L2 English speech from three different native language groups: Italian (9 dyads), French (10 dyads) and Slovak (10 dyads). Our results indicate that the Siamese RNN model effectively captures the dynamics of phonetic convergence and the speaker's imitation ability. Moreover, this text-independent model is scalable and capable of handling L1-induced speaker variability.