HyperLLaVA: Dynamic Visual and Language Expert Tuning for Multimodal Large Language Models

Recent advancements indicate that scaling up Multimodal Large Language Models (MLLMs) effectively enhances performance on downstream multimodal tasks. The prevailing MLLM paradigm, \emph{e.g.}, LLaVA, transforms visual features into text-like tokens using a \emph{static} vision-language mapper, thereby enabling \emph{static} LLMs to develop the capability to comprehend visual information through visual instruction tuning. Although promising, the \emph{static} tuning strategy~\footnote{The static tuning refers to the trained model with static parameters.} that shares the same parameters may constrain performance across different downstream multimodal tasks. In light of this, we introduce HyperLLaVA, which involves adaptive tuning of the projector and LLM parameters, in conjunction with a dynamic visual expert and language expert, respectively. These experts are derived from HyperNetworks, which generates adaptive parameter shifts through visual and language guidance, enabling dynamic projector and LLM modeling in two-stage training. Our experiments demonstrate that our solution significantly surpasses LLaVA on existing MLLM benchmarks, including MME, MMBench, SEED-Bench, and LLaVA-Bench. ~\footnote{Our project is available on the link https://github.com/DCDmllm/HyperLLaVA}.

An Analysis on Matching Mechanisms and Token Pruning for Late-interaction Models

With the development of pre-trained language models, the dense retrieval models have become promising alternatives to the traditional retrieval models that rely on exact match and sparse bag-of-words representations. Different from most dense retrieval models using a bi-encoder to encode each query or document into a dense vector, the recently proposed late-interaction multi-vector models (i.e., ColBERT and COIL) achieve state-of-the-art retrieval effectiveness by using all token embeddings to represent documents and queries and modeling their relevance with a sum-of-max operation. However, these fine-grained representations may cause unacceptable storage overhead for practical search systems. In this study, we systematically analyze the matching mechanism of these late-interaction models and show that the sum-of-max operation heavily relies on the co-occurrence signals and some important words in the document. Based on these findings, we then propose several simple document pruning methods to reduce the storage overhead and compare the effectiveness of different pruning methods on different late-interaction models. We also leverage query pruning methods to further reduce the retrieval latency. We conduct extensive experiments on both in-domain and out-domain datasets and show that some of the used pruning methods can significantly improve the efficiency of these late-interaction models without substantially hurting their retrieval effectiveness.

Large-Scale RIS Enabled Air-Ground Channels: Near-Field Modeling and Analysis

Existing works mainly rely on the far-field planar-wave-based channel model to assess the performance of reconfigurable intelligent surface (RIS)-enabled wireless communication systems. However, when the transmitter and receiver are in near-field ranges, this will result in relatively low computing accuracy. To tackle this challenge, we initially develop an analytical framework for sub-array partitioning. This framework divides the large-scale RIS array into multiple sub-arrays, effectively reducing modeling complexity while maintaining acceptable accuracy. Then, we develop a beam domain channel model based on the proposed sub-array partition framework for large-scale RIS-enabled UAV-to-vehicle communication systems, which can be used to efficiently capture the sparse features in RIS-enabled UAV-to-vehicle channels in both near-field and far-field ranges. Furthermore, some important propagation characteristics of the proposed channel model, including the spatial cross-correlation functions (CCFs), temporal auto-correlation functions (ACFs), frequency correlation functions (CFs), and channel capacities with respect to the different physical features of the RIS and non-stationary properties of the channel model are derived and analyzed. Finally, simulation results are provided to demonstrate that the proposed framework is helpful to achieve a good tradeoff between model complexity and accuracy for investigating the channel propagation characteristics, and therefore providing highly-efficient communications in RIS-enabled UAV-to-vehicle wireless networks.

Human Activity Recognition with Low-Resolution Infrared Array Sensor Using Semi-supervised Cross-domain Neural Networks for Indoor Environment

Low-resolution infrared-based human activity recognition (HAR) attracted enormous interests due to its low-cost and private. In this paper, a novel semi-supervised crossdomain neural network (SCDNN) based on 8 $\times$ 8 low-resolution infrared sensor is proposed for accurately identifying human activity despite changes in the environment at a low-cost. The SCDNN consists of feature extractor, domain discriminator and label classifier. In the feature extractor, the unlabeled and minimal labeled target domain data are trained for domain adaptation to achieve a mapping of the source domain and target domain data. The domain discriminator employs the unsupervised learning to migrate data from the source domain to the target domain. The label classifier obtained from training the source domain data improves the recognition of target domain activities due to the semi-supervised learning utilized in training the target domain data. Experimental results show that the proposed method achieves 92.12\% accuracy for recognition of activities in the target domain by migrating the source and target domains. The proposed approach adapts superior to cross-domain scenarios compared to the existing deep learning methods, and it provides a low-cost yet highly adaptable solution for cross-domain scenarios.

PowerSkel: A Device-Free Framework Using CSI Signal for Human Skeleton Estimation in Power Station

Safety monitoring of power operations in power stations is crucial for preventing accidents and ensuring stable power supply. However, conventional methods such as wearable devices and video surveillance have limitations such as high cost, dependence on light, and visual blind spots. WiFi-based human pose estimation is a suitable method for monitoring power operations due to its low cost, device-free, and robustness to various illumination conditions.In this paper, a novel Channel State Information (CSI)-based pose estimation framework, namely PowerSkel, is developed to address these challenges. PowerSkel utilizes self-developed CSI sensors to form a mutual sensing network and constructs a CSI acquisition scheme specialized for power scenarios. It significantly reduces the deployment cost and complexity compared to the existing solutions. To reduce interference with CSI in the electricity scenario, a sparse adaptive filtering algorithm is designed to preprocess the CSI. CKDformer, a knowledge distillation network based on collaborative learning and self-attention, is proposed to extract the features from CSI and establish the mapping relationship between CSI and keypoints. The experiments are conducted in a real-world power station, and the results show that the PowerSkel achieves high performance with a PCK@50 of 96.27%, and realizes a significant visualization on pose estimation, even in dark environments. Our work provides a novel low-cost and high-precision pose estimation solution for power operation.

Sense-Then-Train: A Novel Beam Training Design for Near-Field MIMO Systems

A novel sense-then-train (STT) scheme is proposed for beam training in near-field multiple-input multiple-output (MIMO) systems. Compared to conventional codebook-based schemes, the proposed STT scheme is capable of not only addressing the complex spherical-wave propagation but also effectively exploiting the additional degrees-of-freedoms (DoFs). The STT scheme is tailored for both single-beam and multi-beam cases. 1) For the single-beam case, the STT scheme first utilizes a sensing phase to estimate a low-dimensional representation of the near-field MIMO channel in the wavenumber domain. Then, in the subsequent training phase, an online learning algorithm is proposed to obtain the optimal beam pair without predefined codebooks or training datasets. 2) For the multi-beam case, based on the single-beam STT, a Gram-Schmidt method is further utilized to guarantee the orthogonality between beams in the training phase. Numerical results unveil that 1) the proposed STT scheme can significantly enhance the beam training performance in the near field compared to the conventional far-field codebook-based schemes, and 2) the proposed STT scheme can perform fast and low-complexity beam training, while achieving a near-optimal performance without full channel state information in both cases.

Video-LaVIT: Unified Video-Language Pre-training with Decoupled Visual-Motional Tokenization

In light of recent advances in multimodal Large Language Models (LLMs), there is increasing attention to scaling them from image-text data to more informative real-world videos. Compared to static images, video poses unique challenges for effective large-scale pre-training due to the modeling of its spatiotemporal dynamics. In this paper, we address such limitations in video-language pre-training with an efficient video decomposition that represents each video as keyframes and temporal motions. These are then adapted to an LLM using well-designed tokenizers that discretize visual and temporal information as a few tokens, thus enabling unified generative pre-training of videos, images, and text. At inference, the generated tokens from the LLM are carefully recovered to the original continuous pixel space to create various video content. Our proposed framework is both capable of comprehending and generating image and video content, as demonstrated by its competitive performance across 13 multimodal benchmarks in image and video understanding and generation. Our code and models will be available at https://video-lavit.github.io.

SD-MVS: Segmentation-Driven Deformation Multi-View Stereo with Spherical Refinement and EM optimization

In this paper, we introduce Segmentation-Driven Deformation Multi-View Stereo (SD-MVS), a method that can effectively tackle challenges in 3D reconstruction of textureless areas. We are the first to adopt the Segment Anything Model (SAM) to distinguish semantic instances in scenes and further leverage these constraints for pixelwise patch deformation on both matching cost and propagation. Concurrently, we propose a unique refinement strategy that combines spherical coordinates and gradient descent on normals and pixelwise search interval on depths, significantly improving the completeness of reconstructed 3D model. Furthermore, we adopt the Expectation-Maximization (EM) algorithm to alternately optimize the aggregate matching cost and hyperparameters, effectively mitigating the problem of parameters being excessively dependent on empirical tuning. Evaluations on the ETH3D high-resolution multi-view stereo benchmark and the Tanks and Temples dataset demonstrate that our method can achieve state-of-the-art results with less time consumption.

Hybrid Vector Message Passing for Generalized Bilinear Factorization

In this paper, we propose a new message passing algorithm that utilizes hybrid vector message passing (HVMP) to solve the generalized bilinear factorization (GBF) problem. The proposed GBF-HVMP algorithm integrates expectation propagation (EP) and variational message passing (VMP) via variational free energy minimization, yielding tractable Gaussian messages. Furthermore, GBF-HVMP enables vector/matrix variables rather than scalar ones in message passing, resulting in a loop-free Bayesian network that improves convergence. Numerical results show that GBF-HVMP significantly outperforms state-of-the-art methods in terms of NMSE performance and computational complexity.