Deep learning models are modern tools for spatio-temporal graph (STG) forecasting. Despite their effectiveness, they require large-scale datasets to achieve better performance and are vulnerable to noise perturbation. To alleviate these limitations, an intuitive idea is to use the popular data augmentation and contrastive learning techniques. However, existing graph contrastive learning methods cannot be directly applied to STG forecasting due to three reasons. First, we empirically discover that the forecasting task is unable to benefit from the pretrained representations derived from contrastive learning. Second, data augmentations that are used for defeating noise are less explored for STG data. Third, the semantic similarity of samples has been overlooked. In this paper, we propose a Spatio-Temporal Graph Contrastive Learning framework (STGCL) to tackle these issues. Specifically, we improve the performance by integrating the forecasting loss with an auxiliary contrastive loss rather than using a pretrained paradigm. We elaborate on four types of data augmentations, which disturb data in terms of graph structure, time domain, and frequency domain. We also extend the classic contrastive loss through a rule-based strategy that filters out the most semantically similar negatives. Our framework is evaluated across three real-world datasets and four state-of-the-art models. The consistent improvements demonstrate that STGCL can be used as an off-the-shelf plug-in for existing deep models.
Given input images, scene graph generation (SGG) aims to produce comprehensive, graphical representations describing visual relationships among salient objects. Recently, more efforts have been paid to the long tail problem in SGG; however, the imbalance in the fraction of missing labels of different classes, or reporting bias, exacerbating the long tail is rarely considered and cannot be solved by the existing debiasing methods. In this paper we show that, due to the missing labels, SGG can be viewed as a "Learning from Positive and Unlabeled data" (PU learning) problem, where the reporting bias can be removed by recovering the unbiased probabilities from the biased ones by utilizing label frequencies, i.e., the per-class fraction of labeled, positive examples in all the positive examples. To obtain accurate label frequency estimates, we propose Dynamic Label Frequency Estimation (DLFE) to take advantage of training-time data augmentation and average over multiple training iterations to introduce more valid examples. Extensive experiments show that DLFE is more effective in estimating label frequencies than a naive variant of the traditional estimate, and DLFE significantly alleviates the long tail and achieves state-of-the-art debiasing performance on the VG dataset. We also show qualitatively that SGG models with DLFE produce prominently more balanced and unbiased scene graphs.
Detecting human-object interactions (HOI) is an important step toward a comprehensive visual understanding of machines. While detecting non-temporal HOIs (e.g., sitting on a chair) from static images is feasible, it is unlikely even for humans to guess temporal-related HOIs (e.g., opening/closing a door) from a single video frame, where the neighboring frames play an essential role. However, conventional HOI methods operating on only static images have been used to predict temporal-related interactions, which is essentially guessing without temporal contexts and may lead to sub-optimal performance. In this paper, we bridge this gap by detecting video-based HOIs with explicit temporal information. We first show that a naive temporal-aware variant of a common action detection baseline does not work on video-based HOIs due to a feature-inconsistency issue. We then propose a simple yet effective architecture named Spatial-Temporal HOI Detection (ST-HOI) utilizing temporal information such as human and object trajectories, correctly-localized visual features, and spatial-temporal masking pose features. We construct a new video HOI benchmark dubbed VidHOI where our proposed approach serves as a solid baseline.
While there has been significant progress towards modelling coherence in written discourse, the work in modelling spoken discourse coherence has been quite limited. Unlike the coherence in text, coherence in spoken discourse is also dependent on the prosodic and acoustic patterns in speech. In this paper, we model coherence in spoken discourse with audio-based coherence models. We perform experiments with four coherence-related tasks with spoken discourses. In our experiments, we evaluate machine-generated speech against the speech delivered by expert human speakers. We also compare the spoken discourses generated by human language learners of varying language proficiency levels. Our results show that incorporating the audio modality along with the text benefits the coherence models in performing downstream coherence related tasks with spoken discourses.
In the era of MOOCs, online exams are taken by millions of candidates, where scoring short answers is an integral part. It becomes intractable to evaluate them by human graders. Thus, a generic automated system capable of grading these responses should be designed and deployed. In this paper, we present a fast, scalable, and accurate approach towards automated Short Answer Scoring (SAS). We propose and explain the design and development of a system for SAS, namely AutoSAS. Given a question along with its graded samples, AutoSAS can learn to grade that prompt successfully. This paper further lays down the features such as lexical diversity, Word2Vec, prompt, and content overlap that plays a pivotal role in building our proposed model. We also present a methodology for indicating the factors responsible for scoring an answer. The trained model is evaluated on an extensively used public dataset, namely Automated Student Assessment Prize Short Answer Scoring (ASAP-SAS). AutoSAS shows state-of-the-art performance and achieves better results by over 8% in some of the question prompts as measured by Quadratic Weighted Kappa (QWK), showing performance comparable to humans.
In this work, we explore a new problem of frame interpolation for speech videos. Such content today forms the major form of online communication. We try to solve this problem by using several deep learning video generation algorithms to generate the missing frames. We also provide examples where computer vision models despite showing high performance on conventional non-linguistic metrics fail to accurately produce faithful interpolation of speech. With this motivation, we provide a new set of linguistically-informed metrics specifically targeted to the problem of speech videos interpolation. We also release several datasets to test computer vision video generation models of their speech understanding.
Domain divergence plays a significant role in estimating the performance of a model when applied to new domains. While there is significant literature on divergence measures, choosing an appropriate divergence measures remains difficult for researchers. We address this shortcoming by both surveying the literature and through an empirical study. We contribute a taxonomy of divergence measures consisting of three groups -- Information-theoretic, Geometric, and Higher-order measures -- and identify the relationships between them. We then ground the use of divergence measures in three different application groups -- 1) Data Selection, 2) Learning Representation, and 3) Decisions in the Wild. From this, we identify that Information-theoretic measures are prevalent for 1) and 3), and higher-order measures are common for 2). To further help researchers, we validate these uses empirically through a correlation analysis of performance drops. We consider the current contextual word representations (CWR) to contrast with the older word distribution based representations for this analysis. We find that traditional measures over word distributions still serve as strong baselines, while higher-order measures with CWR are effective.
Deep Learning and its applications have cascaded impactful research and development with a diverse range of modalities present in the real-world data. More recently, this has enhanced research interests in the intersection of the Vision and Language arena with its numerous applications and fast-paced growth. In this paper, we present a detailed overview of the latest trends in research pertaining to visual and language modalities. We look at its applications in their task formulations and how to solve various problems related to semantic perception and content generation. We also address task-specific trends, along with their evaluation strategies and upcoming challenges. Moreover, we shed some light on multi-disciplinary patterns and insights that have emerged in the recent past, directing this field towards more modular and transparent intelligent systems. This survey identifies key trends gravitating recent literature in VisLang research and attempts to unearth directions that the field is heading towards.
Visual relationship detection aims to reason over relationships among salient objects in images, which has drawn increasing attention over the past few years. Inspired by human reasoning mechanism, it is believed that external visual commonsense knowledge is beneficial for reasoning visual relationships of objects in images, which is however rarely considered in existing methods. In this paper, we propose a novel approach named Relational Visual-Linguistic Bidirectional Encoder Representations from Transformers (RVL-BERT), which performs relational reasoning with both visual and language commonsense knowledge learned via self-supervised pre-training with multimodal representations. RVL-BERT also uses an effective spatial module and a novel mask attention module to explicitly capture spatial information among the objects. Moreover, our model decouples object detection from visual relationship recognition by taking in object names directly, enabling it to be used on top of any object detection system. We show through quantitative and qualitative experiments that, with the transferred knowledge and novel modules, RVL-BERT surpasses previous state-of-the-art on two challenging visual relationship detection datasets. The source code will be publicly available soon.