AAAI 2024 Papers

Omnidirectional images have attracted significant attention in recent years due to the rapid development of virtual reality technologies. Equirectangular projection (ERP), a naive form to store and transfer omnidirectional images, however, is challenging for existing two-dimensional (2D) image super-resolution (SR) methods due to its inhomogeneous distributed sampling density and distortion across latitude. In this paper, we make one of the first attempts to design a spherical pseudo-cylindrical representation, which not only allows pixels at different latitudes to adaptively adopt the best distinct sampling density but also is model-agnostic to most off-the-shelf SR methods, enhancing their performances. Specifically, we start by upsampling each latitude of the input ERP image and design a computationally tractable optimization algorithm to adaptively obtain a (sub)-optimal sampling density for each latitude of the ERP image. Addressing the distortion of ERP, we introduce a new viewport-based training loss based on the original 3D sphere format of the omnidirectional image, which inherently lacks distortion. Finally, we present a simple yet effective recursive progressive omnidirectional SR network to showcase the feasibility of our idea. The experimental results on public datasets demonstrate the effectiveness of the proposed method as well as the consistently superior performance of our method over most state-of-the-art methods both quantitatively and qualitatively.

Controllable spherical panoramic image generation holds substantial applicative potential across a variety of domains. However, it remains a challenging task due to the inherent spherical distortion and geometry characteristics, resulting in low-quality content generation. In this paper, we introduce a novel framework of SphereDiffusion to address these unique challenges, for better generating high-quality and precisely controllable spherical panoramic images. For the spherical distortion characteristic, we embed the semantics of the distorted object with text encoding, then explicitly construct the relationship with text-object correspondence to better use the pre-trained knowledge of the planar images. Meanwhile, we employ a deformable technique to mitigate the semantic deviation in latent space caused by spherical distortion. For the spherical geometry characteristic, in virtue of spherical rotation invariance, we improve the data diversity and optimization objectives in the training process, enabling the model to better learn the spherical geometry characteristic. Furthermore, we enhance the denoising process of the diffusion model, enabling it to effectively use the learned geometric characteristic to ensure the boundary continuity of the generated images. With these specific techniques, experiments on Structured3D dataset show that SphereDiffusion significantly improves the quality of controllable spherical image generation and relatively reduces around 35% FID on average.

Currently, image generation and synthesis have remarkably progressed with generative models. Despite photo-realistic results, intrinsic discrepancies are still observed in the frequency domain. The spectral discrepancy appeared not only in generative adversarial networks but in diffusion models. In this study, we propose a framework to effectively mitigate the disparity in frequency domain of the generated images to improve generative performance of both GAN and diffusion models. This is realized by spectrum translation for the refinement of image generation (STIG) based on contrastive learning. We adopt theoretical logic of frequency components in various generative networks. The key idea, here, is to refine the spectrum of the generated image via the concept of image-to-image translation and contrastive learning in terms of digital signal processing. We evaluate our framework across eight fake image datasets and various cutting-edge models to demonstrate the effectiveness of STIG. Our framework outperforms other cutting-edges showing significant decreases in FID and log frequency distance of spectrum. We further emphasize that STIG improves image quality by decreasing the spectral anomaly. Additionally, validation results present that the frequency-based deepfake detector confuses more in the case where fake spectrums are manipulated by STIG.

In this paper, we propose SpectralNeRF, an end-to-end Neural Radiance Field (NeRF)-based architecture for high-quality physically based rendering from a novel spectral perspective. We modify the classical spectral rendering into two main steps, 1) the generation of a series of spectrum maps spanning different wavelengths, 2) the combination of these spectrum maps for the RGB output. Our SpectralNeRF follows these two steps through the proposed multi-layer perceptron (MLP)-based architecture (SpectralMLP) and Spectrum Attention UNet (SAUNet). Given the ray origin and the ray direction, the SpectralMLP constructs the spectral radiance field to obtain spectrum maps of novel views, which are then sent to the SAUNet to produce RGB images of white-light illumination. Applying NeRF to build up the spectral rendering is a more physically-based way from the perspective of ray-tracing. Further, the spectral radiance fields decompose difficult scenes and improve the performance of NeRF-based methods. Comprehensive experimental results demonstrate the proposed SpectralNeRF is superior to recent NeRF-based methods when synthesizing new views on synthetic and real datasets. The codes and datasets are available at https://github.com/liru0126/SpectralNeRF.

Modeling complementary relationships greatly helps recommender systems to accurately and promptly recommend the subsequent items when one item is purchased. Unlike traditional similar relationships, items with complementary relationships may be purchased successively (such as iPhone and Airpods Pro), and they not only share relevance but also exhibit dissimilarity. Since the two attributes are opposites, modeling complementary relationships is challenging. Previous attempts to exploit these relationships have either ignored or oversimplified the dissimilarity attribute, resulting in ineffective modeling and an inability to balance the two attributes. Since Graph Neural Networks (GNNs) can capture the relevance and dissimilarity between nodes in the spectral domain, we can leverage spectral-based GNNs to effectively understand and model complementary relationships. In this study, we present a novel approach called Spectral-based Complementary Graph Neural Networks (SComGNN) that utilizes the spectral properties of complementary item graphs. We make the first observation that complementary relationships consist of low-frequency and mid-frequency components, corresponding to the relevance and dissimilarity attributes, respectively. Based on this spectral observation, we design spectral graph convolutional networks with low-pass and mid-pass filters to capture the low-frequency and mid-frequency components. Additionally, we propose a two-stage attention mechanism to adaptively integrate and balance the two attributes. Experimental results on four e-commerce datasets demonstrate the effectiveness of our model, with SComGNN significantly outperforming existing baseline models.

Recently, CLIP has found practical utility in the domain of pixel-level zero-shot segmentation tasks. The present landscape features two-stage methodologies beset by issues such as intricate pipelines and elevated computational costs. While current one-stage approaches alleviate these concerns and incorporate Visual Prompt Training (VPT) to uphold CLIP's generalization capacity, they still fall short in fully harnessing CLIP's potential for pixel-level unseen class demarcation and precise pixel predictions. To further stimulate CLIP's zero-shot dense prediction capability, we propose SPT-SEG, a one-stage approach that improves CLIP's adaptability from image to pixel. Specifically, we initially introduce Spectral Prompt Tuning (SPT), incorporating spectral prompts into the CLIP visual encoder's shallow layers to capture structural intricacies of images, thereby enhancing comprehension of unseen classes. Subsequently, we introduce the Spectral Guided Decoder (SGD), utilizing both high and low-frequency information to steer the network's spatial focus towards more prominent classification features, enabling precise pixel-level prediction outcomes. Through extensive experiments on two public datasets, we demonstrate the superiority of our method over state-of-the-art approaches, performing well across all classes and particularly excelling in handling unseen classes.

Real-world graphs are dynamic, constantly evolving with new interactions, such as financial transactions in financial networks. Temporal Graph Neural Networks (TGNNs) have been developed to effectively capture the evolving patterns in dynamic graphs. While these models have demonstrated their superiority, being widely adopted in various important fields, their vulnerabilities against adversarial attacks remain largely unexplored. In this paper, we propose T-SPEAR, a simple and effective adversarial attack method for link prediction on continuous-time dynamic graphs, focusing on investigating the vulnerabilities of TGNNs. Specifically, before the training procedure of a victim model, which is a TGNN for link prediction, we inject edge perturbations to the data that are unnoticeable in terms of the four constraints we propose, and yet effective enough to cause malfunction of the victim model. Moreover, we propose a robust training approach T-SHIELD to mitigate the impact of adversarial attacks. By using edge filtering and enforcing temporal smoothness to node embeddings, we enhance the robustness of the victim model. Our experimental study shows that T-SPEAR significantly degrades the victim model's performance on link prediction tasks, and even more, our attacks are transferable to other TGNNs, which differ from the victim model assumed by the attacker. Moreover, we demonstrate that T-SHIELD effectively filters out adversarial edges and exhibits robustness against adversarial attacks, surpassing the link prediction performance of the naive TGNN by up to 11.2% under T-SPEAR. The code and datasets are available at https://github.com/wooner49/T-spear-shield

Traffic flow forecasting is a classical spatio-temporal data mining problem with many real-world applications. Recently, various methods based on Graph Neural Networks (GNN) have been proposed for the problem and achieved impressive prediction performance. However, we argue that the majority of existing methods disregarding the importance of certain nodes (referred to as pivotal nodes) that naturally exhibit extensive connections with multiple other nodes. Predicting on pivotal nodes poses a challenge due to their complex spatio-temporal dependencies compared to other nodes. In this paper, we propose a novel GNN-based method called Spatio-Temporal Pivotal Graph Neural Networks (STPGNN) to address the above limitation. We introduce a pivotal node identification module for identifying pivotal nodes. We propose a novel pivotal graph convolution module, enabling precise capture of spatio-temporal dependencies centered around pivotal nodes. Moreover, we propose a parallel framework capable of extracting spatio-temporal traffic features on both pivotal and non-pivotal nodes. Experiments on seven real-world traffic datasets verify our proposed method's effectiveness and efficiency compared to state-of-the-art baselines.

Graph Convolutional Networks (GCNs) and Transformers have been widely applied to skeleton-based human action recognition, with each offering unique advantages in capturing spatial relationships and long-range dependencies. However, for most GCN methods, the construction of topological structures relies solely on the spatial information of human joints, limiting their ability to directly capture richer spatio-temporal dependencies. Additionally, the self-attention modules of many Transformer methods lack topological structure information, restricting the robustness and generalization of the models. To address these issues, we propose a Joint Trajectory Graph (JTG) that integrates spatio-temporal information into a uniform graph structure. We also present a Joint Trajectory GraphFormer (JT-GraphFormer), which directly captures the spatio-temporal relationships among all joint trajectories for human action recognition. To better integrate topological information into spatio-temporal relationships, we introduce a Spatio-Temporal Dijkstra Attention (STDA) mechanism to calculate relationship scores for all the joints in JTG. Furthermore, we incorporate the Koopman operator into the classification stage to enhance the model's representation ability and classification performance. Experiments demonstrate that JT-GraphFormer achieves outstanding performance in human action recognition tasks, outperforming state-of-the-art methods on the NTU RGB+D, NTU RGB+D 120, and N-UCLA datasets.

In the realm of human mobility, the decision-making process for selecting the next-visit location is intricately influenced by a trade-off between spatial and temporal constraints, which are reflective of individual needs and preferences. This trade-off, however, varies across individuals, making the modeling of these spatial-temporal dynamics a formidable challenge. To address the problem, in this work, we introduce the "Spatial-temporal Induced Hierarchical Reinforcement Learning" (STI-HRL) framework, for capturing the interplay between spatial and temporal factors in human mobility decision-making. Specifically, STI-HRL employs a two-tiered decision-making process: the low-level focuses on disentangling spatial and temporal preferences using dedicated agents, while the high-level integrates these considerations to finalize the decision. To complement the hierarchical decision setting, we construct a hypergraph to organize historical data, encapsulating the multi-aspect semantics of human mobility. We propose a cross-channel hypergraph embedding module to learn the representations as the states to facilitate the decision-making cycle. Our extensive experiments on two real-world datasets validate the superiority of STI-HRL over state-of-the-art methods in predicting users' next visits across various performance metrics.

Crime prediction stands as a pivotal concern within the realm of urban management due to its potential threats to public safety. While prior research has predominantly focused on unraveling the intricate dependencies among urban regions and temporal dynamics, the challenges posed by the scarcity and uncertainty of historical crime data have not been thoroughly investigated. This study introduces an innovative spatial-temporal augmented learning framework for crime prediction, namely STAug. In STAug, we devise a CrimeMix to improve the ability of generalization. Furthermore, we harness a spatial-temporal aggregation to capture and incorporate multiple correlations covering the temporal, spatial, and crime-type aspects. Experiments on two real-world datasets underscore the superiority of STAug over several baselines.

A large amount of User Generated Content (UGC) is uploaded to the Internet daily and displayed to people world-widely through the client side (mobile and PC). This requires the cropping algorithms to produce the aesthetic thumbnail within a specific aspect ratio on different devices. However, existing image cropping works mainly focus on landmark or landscape images, which fail to model the relations among the multi-objects with the complex background in UGC. Besides, previous methods merely consider the aesthetics of the cropped images while ignoring the content integrity, which is crucial for UGC cropping. In this paper, we propose a Spatial-Semantic Collaborative cropping network (S2CNet) for arbitrary user generated content accompanied by a new cropping benchmark. Specifically, we first mine the visual genes of the potential objects. Then, the suggested adaptive attention graph recasts this task as a procedure of information association over visual nodes. The underlying spatial and semantic relations are ultimately centralized to the crop candidate through differentiable message passing, which helps our network efficiently to preserve both the aesthetics and the content integrity. Extensive experiments on the proposed UGCrop5K and other public datasets demonstrate the superiority of our approach over state-of-the-art counterparts.

Leveraging wearable devices for motion reconstruction has emerged as an economical and viable technique. Certain methodologies employ sparse Inertial Measurement Units (IMUs) on the human body and harness data-driven strategies to model human poses. However, the reconstruction of motion based solely on sparse IMU data is inherently fraught with ambiguity, a consequence of numerous identical IMU readings corresponding to different poses. In this paper, we explore the spatial importance of sparse sensors, supervised by text that describes specific actions. Specifically, uncertainty is introduced to derive weighted features for each IMU. We also design a Hierarchical Temporal Transformer (HTT) and apply contrastive learning to achieve precise temporal and feature alignment of sensor data with textual semantics. Experimental results demonstrate our proposed approach achieves significant improvements in multiple metrics compared to existing methods. Notably, with textual supervision, our method not only differentiates between ambiguous actions such as sitting and standing but also produces more precise and natural motion.

Flood mapping on Earth imagery is crucial for disaster management, but its efficacy is hampered by the lack of high-quality training labels. Given high-resolution Earth imagery with coarse and noisy training labels, a base deep neural network model, and a spatial knowledge base with label constraints, our problem is to infer the true high-resolution labels while training neural network parameters. Traditional methods are largely based on specific physical properties and thus fall short of capturing the rich domain constraints expressed by symbolic logic. Neural-symbolic models can capture rich domain knowledge, but existing methods do not address the unique spatial challenges inherent in flood mapping on high-resolution imagery. To fill this gap, we propose a spatial-logic-aware weakly supervised learning framework. Our framework integrates symbolic spatial logic inference into probabilistic learning in a weakly supervised setting. To reduce the time costs of logic inference on vast high-resolution pixels, we propose a multi-resolution spatial reasoning algorithm to infer true labels while training neural network parameters. Evaluations of real-world flood datasets show that our model outperforms several baselines in prediction accuracy. The code is available at https://github.com/spatialdatasciencegroup/SLWSL.

Most existing GAN inversion methods either achieve accurate reconstruction but lack editability or offer strong editability at the cost of fidelity. Hence, how to balance the distortion-editability trade-off is a significant challenge for GAN inversion. To address this challenge, we introduce a novel spatial-contextual discrepancy information compensation-based GAN-inversion method (SDIC), which consists of a discrepancy information prediction network (DIPN) and a discrepancy information compensation network (DICN). SDIC follows a ``compensate-and-edit'' paradigm and successfully bridges the gap in image details between the original image and the reconstructed/edited image. On the one hand, DIPN encodes the multi-level spatial-contextual information of the original and initial reconstructed images and then predicts a spatial-contextual guided discrepancy map with two hourglass modules. In this way, a reliable discrepancy map that models the contextual relationship and captures fine-grained image details is learned. On the other hand, DICN incorporates the predicted discrepancy information into both the latent code and the GAN generator with different transformations, generating high-quality reconstructed/edited images. This effectively compensates for the loss of image details during GAN inversion. Both quantitative and qualitative experiments demonstrate that our proposed method achieves the excellent distortion-editability trade-off at a fast inference speed for both image inversion and editing tasks. Our code is available at https://github.com/ZzqLKED/SDIC.

We consider spatial voting where candidates are located in the Euclidean d-dimensional space, and each voter ranks candidates based on their distance from the voter's ideal point. We explore the case where information about the location of voters' ideal points is incomplete: for each dimension, we are given an interval of possible values. We study the computational complexity of finding the possible and necessary winners for positional scoring rules. Our results show that we retain tractable cases of the classic model where voters have partial-order preferences. Moreover, we show that there are positional scoring rules under which the possible-winner problem is intractable for partial orders, but tractable in the one-dimensional spatial setting. We also consider approval voting in this setting. We show that for up to two dimensions, the necessary-winner problem is tractable, while the possible-winner problem is hard for any number of dimensions.

Vision Transformers (ViTs) have achieved remarkable success in computer vision tasks. However, their potential in rotation-sensitive scenarios has not been fully explored, and this limitation may be inherently attributed to the lack of spatial invariance in the data-forwarding process. In this study, we present a novel approach, termed Spatial Transform Decoupling (STD), providing a simple-yet-effective solution for oriented object detection with ViTs. Built upon stacked ViT blocks, STD utilizes separate network branches to predict the position, size, and angle of bounding boxes, effectively harnessing the spatial transform potential of ViTs in a divide-and-conquer fashion. Moreover, by aggregating cascaded activation masks (CAMs) computed upon the regressed parameters, STD gradually enhances features within regions of interest (RoIs), which complements the self-attention mechanism. Without bells and whistles, STD achieves state-of-the-art performance on the benchmark datasets including DOTA-v1.0 (82.24% mAP) and HRSC2016 (98.55% mAP), which demonstrates the effectiveness of the proposed method. Source code is available at https://github.com/yuhongtian17/Spatial-Transform-Decoupling.

Large Language Models (LLMs) have achieved unprecedented breakthroughs in various natural language processing domains. However, the enigmatic ``black-box'' nature of LLMs remains a significant challenge for interpretability, hampering transparent and accountable applications. While past approaches, such as attention visualization, pivotal subnetwork extraction, and concept-based analyses, offer some insight, they often focus on either local or global explanations within a single dimension, occasionally falling short in providing comprehensive clarity. In response, we propose a novel methodology anchored in sparsity-guided techniques, aiming to provide a holistic interpretation of LLMs. Our framework, termed SparseCBM, innovatively integrates sparsity to elucidate three intertwined layers of interpretation: input, subnetwork, and concept levels. In addition, the newly introduced dimension of interpretable inference-time intervention facilitates dynamic adjustments to the model during deployment. Through rigorous empirical evaluations on real-world datasets, we demonstrate that SparseCBM delivers a profound understanding of LLM behaviors, setting it apart in both interpreting and ameliorating model inaccuracies. Codes are provided in supplements.

We study to generate novel views of indoor scenes given sparse input views. The challenge is to achieve both photorealism and view consistency. We present SparseGNV: a learning framework that incorporates 3D structures and image generative models to generate novel views with three modules. The first module builds a neural point cloud as underlying geometry, providing scene context and guidance for the target novel view. The second module utilizes a transformer-based network to map the scene context and the guidance into a shared latent space and autoregressively decodes the target view in the form of discrete image tokens. The third module reconstructs the tokens back to the image of the target view. SparseGNV is trained across a large-scale indoor scene dataset to learn generalizable priors. Once trained, it can efficiently generate novel views of an unseen indoor scene in a feed-forward manner. We evaluate SparseGNV on real-world indoor scenes and demonstrate that it outperforms state-of-the-art methods based on either neural radiance fields or conditional image generation.