Neural Radiance Fields (NeRF) often struggle with reconstructing and rendering highly reflective scenes. Recent advancements have developed various reflection-aware appearance models to enhance NeRF's capability to render specular reflections. However, the robust reconstruction of highly reflective scenes is still hindered by the inherent shape ambiguity on specular surfaces. Existing methods typically rely on additional geometry priors to regularize the shape prediction, but this can lead to oversmoothed geometry in complex scenes. Observing the critical role of surface normals in parameterizing reflections, we introduce a transmittance-gradient-based normal estimation technique that remains robust even under ambiguous shape conditions. Furthermore, we propose a dual activated densities module that effectively bridges the gap between smooth surface normals and sharp object boundaries. Combined with a reflection-aware appearance model, our proposed method achieves robust reconstruction and high-fidelity rendering of scenes featuring both highly specular reflections and intricate geometric structures. Extensive experiments demonstrate that our method outperforms existing state-of-the-art methods on various datasets.

In this paper, we propose a new multi-modal task, termed audio-visual instance segmentation (AVIS), which aims to simultaneously identify, segment and track individual sounding object instances in audible videos. To facilitate this research, we introduce a high-quality benchmark named AVISeg, containing over 90K instance masks from 26 semantic categories in 926 long videos. Additionally, we propose a strong baseline model for this task. Our model first localizes sound source within each frame, and condenses object-specific contexts into concise tokens. Then it builds long-range audio-visual dependencies between these tokens using window-based attention, and tracks sounding objects among the entire video sequences. Extensive experiments reveal that our method performs best on AVISeg, surpassing the existing methods from related tasks. We further conduct the evaluation on several multi-modal large models. Unfortunately, they exhibits subpar performance on instance-level sound source localization and temporal perception. We expect that AVIS will inspire the community towards a more comprehensive multi-modal understanding. Dataset and code is available at https://github.com/ruohaoguo/avis.

This paper presents FreEformer, a simple yet effective model that leverages a Frequency Enhanced Transformer for multivariate time series forecasting. Our work is based on the assumption that the frequency spectrum provides a global perspective on the composition of series across various frequencies and is highly suitable for robust representation learning. Specifically, we first convert time series into the complex frequency domain using the Discrete Fourier Transform (DFT). The Transformer architecture is then applied to the frequency spectra to capture cross-variate dependencies, with the real and imaginary parts processed independently. However, we observe that the vanilla attention matrix exhibits a low-rank characteristic, thus limiting representation diversity. To address this, we enhance the vanilla attention mechanism by introducing an additional learnable matrix to the original attention matrix, followed by row-wise L1 normalization. Theoretical analysis demonstrates that this enhanced attention mechanism improves both feature diversity and gradient flow. Extensive experiments demonstrate that FreEformer consistently outperforms state-of-the-art models on eighteen real-world benchmarks covering electricity, traffic, weather, healthcare and finance. Notably, the enhanced attention mechanism also consistently improves the performance of state-of-the-art Transformer-based forecasters. Code is available at https://anonymous.4open.science/r/FreEformer.

This paper presents $\mathbf{OLinear}$, a $\mathbf{linear}$-based multivariate time series forecasting model that operates in an $\mathbf{o}$rthogonally transformed domain. Recent forecasting models typically adopt the temporal forecast (TF) paradigm, which directly encode and decode time series in the time domain. However, the entangled step-wise dependencies in series data can hinder the performance of TF. To address this, some forecasters conduct encoding and decoding in the transformed domain using fixed, dataset-independent bases (e.g., sine and cosine signals in the Fourier transform). In contrast, we propose $\mathbf{OrthoTrans}$, a data-adaptive transformation based on an orthogonal matrix that diagonalizes the series' temporal Pearson correlation matrix. This approach enables more effective encoding and decoding in the decorrelated feature domain and can serve as a plug-in module to enhance existing forecasters. To enhance the representation learning for multivariate time series, we introduce a customized linear layer, $\mathbf{NormLin}$, which employs a normalized weight matrix to capture multivariate dependencies. Empirically, the NormLin module shows a surprising performance advantage over multi-head self-attention, while requiring nearly half the FLOPs. Extensive experiments on 24 benchmarks and 140 forecasting tasks demonstrate that OLinear consistently achieves state-of-the-art performance with high efficiency. Notably, as a plug-in replacement for self-attention, the NormLin module consistently enhances Transformer-based forecasters. The code and datasets are available at https://github.com/jackyue1994/OLinear.

In the field of vanishing point detection, previous works commonly relied on extracting and clustering straight lines or classifying candidate points as vanishing points. This paper proposes a novel end-to-end framework, called VPDETR (Vanishing Point DEtection TRansformer), that views vanishing point detection as a set prediction problem, applicable to both Manhattan and non-Manhattan world datasets. By using the positional embedding of anchor points as queries in Transformer decoders and dynamically updating them layer by layer, our method is able to directly input images and output their vanishing points without the need for explicit straight line extraction and candidate points sampling. Additionally, we introduce an orthogonal loss and a cross-prediction loss to improve accuracy on the Manhattan world datasets. Experimental results demonstrate that VPDETR achieves competitive performance compared to state-of-the-art methods, without requiring post-processing.

Point cloud is a popular and widely used geometric representation, which has attracted significant attention in 3D vision. However, the geometric variability of point cloud representations across different datasets can cause domain discrepancies, which hinder knowledge transfer and model generalization, resulting in degraded performance in target domain. In this paper, we present a novel approach to improve point cloud domain adaptation by employing masked representation learning in a self-supervised manner. Specifically, our method combines masked feature prediction and masked sample consistency to encode both local structure and global semantic information for learning invariant point cloud representation across domains. Moreover, to learn domain-specific representation and transfer knowledge from source to target, we propose prototype-calibrated self-training. By exploiting class-wise prototypes in the shared feature space, the soft pseudo labels can be adaptively denoised, which benefits the decision boundary learning in target domain. We conduct experiments on PointDA-10 and PointSegDA for 3D point cloud shape classification and semantic segmentation, respectively. The results demonstrate the effectiveness of our method and show that we can achieve the new state-of-the-art performance on point cloud domain adaptation.

Panoramic audio-visual saliency detection is to segment the most attention-attractive regions in 360° panoramic videos with sound. To meticulously delineate the detected salient regions and effectively model human attention shift, we extend this task to more fine-grained instance scenarios: identifying salient object instances and inferring their saliency ranks. In this paper, we propose the first instance-level framework that can simultaneously be applied to segmentation and ranking of multiple salient objects in panoramic videos. Specifically, it consists of a distortion-aware pixel decoder to overcome panoramic distortions, a sequential audio-visual fusion module to integrate audio-visual information, and a spatio-temporal object decoder to separate individual instances and predict their saliency scores. Moreover, owing to the absence of such annotations, we create the ground-truth saliency ranks for the PAVS10K benchmark. Extensive experiments demonstrate that our model is capable of achieving state-of-the-art performance on the PAVS10K for both saliency detection and ranking tasks. The code is available at https://github.com/ruohaoguo/pavsodr.

Audio-visual semantic segmentation (AVSS) aims to segment and classify sounding objects in videos with acoustic cues. However, most approaches operate on the close-set assumption and only identify pre-defined categories from training data, lacking the generalization ability to detect novel categories in practical applications. In this paper, we introduce a new task: open-vocabulary audio-visual semantic segmentation, extending AVSS task to open-world scenarios beyond the annotated label space. This is a more challenging task that requires recognizing all categories, even those that have never been seen nor heard during training. Moreover, we propose the first open-vocabulary AVSS framework, OV-AVSS, which mainly consists of two parts: 1) a universal sound source localization module to perform audio-visual fusion and locate all potential sounding objects and 2) an open-vocabulary classification module to predict categories with the help of the prior knowledge from large-scale pre-trained vision-language models. To properly evaluate the open-vocabulary AVSS, we split zero-shot training and testing subsets based on the AVSBench-semantic benchmark, namely AVSBench-OV. Extensive experiments demonstrate the strong segmentation and zero-shot generalization ability of our model on all categories. On the AVSBench-OV dataset, OV-AVSS achieves 55.43% mIoU on base categories and 29.14% mIoU on novel categories, exceeding the state-of-the-art zero-shot method by 41.88%/20.61% and open-vocabulary method by 10.2%/11.6%. The code is available at https://github.com/ruohaoguo/ovavss.

Exploiting the rich temporal information in human pose sequences to facilitate 3D pose estimation has garnered particular attention. While various learning architectures have been designed for temporal exploiting, these architectures are usually trained via the 3D pose loss independently imposed on every single frame, without explicit temporal signals introduced for supervision. This inevitably increases the difficulty of temporal exploiting, since the network must reason about the meaningful temporal information based on the non-temporal single-frame supervision first. Only then, the network can utilize this information to guide sequence modeling. Recently, some work introduce temporal smoothness as an explicit supervision signal, which makes the network more straightforwardly reaches the temporal information from the supervision signal, thus improving the temporal exploiting. However, the temporal smoothness only roughly measures the short-term temporal properties between adjacent frame pairs. In this work, we propose to generalize the supervision of temporal smoothness to temporal correlations, letting the network precisely consider more comprehensive temporal properties in sequences. We contribute two novel correlation-based loss functions, which adopt different strategies to respectively regularize the encoder and decoder sides of the network for temporal exploiting. Besides, we design a pre-training scheme to ensure a general convergence of existing pose estimators under our correlation losses. Experiments on three benchmarks demonstrate that our method can be compatible with different networks, improving their temporal exploiting ability to output more accurate and robust pose estimations.

In this work, we investigate contrastive learning on perturbed point clouds and find that the contrasting process may widen the domain gap caused by random perturbations, making the pre-trained network fail to generalize on testing data. To this end, we propose the Equivariant COntrastive framework which closes the domain gap before contrasting, further introduces the equivariance property, and enables pre-training networks under more perturbation types to obtain meaningful features. Specifically, to close the domain gap, a pre-trained VAE is adopted to convert perturbed point clouds into less perturbed point embedding of similar domains and separated perturbation embedding. The contrastive pairs can then be generated by mixing the point embedding with different perturbation embedding. Moreover, to pursue the equivariance property, a Vector Quantizer is adopted during VAE training, discretizing the perturbation embedding into one-hot tokens which indicate the perturbation labels. By correctly predicting the perturbation labels from the perturbed point cloud, the property of equivariance can be encouraged in the learned features. Experiments on synthesized and real-world perturbed datasets show that ECO-3D outperforms most existing pre-training strategies under various downstream tasks, achieving SOTA performance for lots of perturbations.

Domain adaptation for 3D point cloud has attracted a lot of interest since it can avoid the time-consuming labeling process of 3D data to some extent. A recent work named xMUDA leveraged multi-modal data to domain adaptation task of 3D semantic segmentation by mimicking the predictions between 2D and 3D modalities, and outperformed the previous single modality methods only using point clouds. Based on it, in this paper, we propose a novel cross-modal contrastive learning scheme to further improve the adaptation effects. By employing constraints from the correspondences between 2D pixel features and 3D point features, our method not only facilitates interaction between the two different modalities, but also boosts feature representations in both labeled source domain and unlabeled target domain. Meanwhile, to sufficiently utilize 2D context information for domain adaptation through cross-modal learning, we introduce a neighborhood feature aggregation module to enhance pixel features. The module employs neighborhood attention to aggregate nearby pixels in the 2D image, which relieves the mismatching between the two different modalities, arising from projecting relative sparse point cloud to dense image pixels. We evaluate our method on three unsupervised domain adaptation scenarios, including country-to-country, day-to-night, and dataset-to-dataset. Experimental results show that our approach outperforms existing methods, which demonstrates the effectiveness of the proposed method.

Deep learning based 3D point cloud classification and segmentation has achieved remarkable success. Existing methods are usually implemented in the original space with 3D coordinates as inputs. However, we find that point networks taking only information of first-order coordinates hardly learn geometric features of higher order, such as point cloud normals or poses. In this study, we propose to map the input point clouds into a non-linear space to facilitate networks learning and leveraging high-order features. Firstly, we design the Parametric Veronese Mapping (PVM) function which automatically learns to map point clouds into a non-linear space. As a result, the mapped point clouds are enriched with high-order elements and maintain the basic point set properties as in the original 3D space. We can then exploit existing networks to learn high-order features from mapped point clouds. Secondly, we contribute a two-stage transformation learning module that modifies the previous one-stage module to better leverage high-order features for aligning point clouds in the projective space. Finally, an interaction module is designed to learn more discriminative features by aggregating information from both the original and projective space. Extensive experiments demonstrate that our method successfully improves the ability of most existing networks to learn high-order features and thus contributing to more accurate classification and segmentation. Moreover, the resulting models show stronger robustness to affine transformations and real-world perturbations.

Knowledge graph embedding aims to represent entities and relations as low-dimensional vectors, which is an effective way for predicting missing links. It is crucial for knowledge graph embedding models to model and infer various relation patterns, such as symmetry/antisymmetry. However, many existing approaches fail to model semantic hierarchies, which are common in the real world. We propose a new model called HRQE, which represents entities as pure quaternions. The relational embedding consists of two parts: (a) Using unit quaternions to represent the rotation part in 3D space, where the head entities are rotated by the corresponding relations through Hamilton product. (b) Using scale parameters to constrain the modulus of entities to make them have hierarchical distributions. To the best of our knowledge, HRQE is the first model that can encode symmetry/antisymmetry, inversion, composition, multiple relation patterns and learn semantic hierarchies simultaneously. Experimental results demonstrate the effectiveness of HRQE against some of the SOTA methods on four well-established knowledge graph completion benchmarks.