Shuo Zhang

Zhang Li, Biao Yang, Qiang Liu et al.

Large Multimodal Models (LMMs) have shown promise in vision-language tasks but struggle with high-resolution input and detailed scene understanding. Addressing these challenges, we introduce Monkey to enhance LMM capabilities. Firstly, Monkey processes input images by dividing them into uniform patches, each matching the size (e.g., 448x448) used in the original training of the well-trained vision encoder. Equipped with individual adapter for each patch, Monkey can handle higher resolutions up to 1344x896 pixels, enabling the detailed capture of complex visual information. Secondly, it employs a multi-level description generation method, enriching the context for scene-object associations. This two-part strategy ensures more effective learning from generated data: the higher resolution allows for a more detailed capture of visuals, which in turn enhances the effectiveness of comprehensive descriptions. Extensive ablative results validate the effectiveness of our designs. Additionally, experiments on 18 datasets further demonstrate that Monkey surpasses existing LMMs in many tasks like Image Captioning and various Visual Question Answering formats. Specially, in qualitative tests focused on dense text question answering, Monkey has exhibited encouraging results compared with GPT4V. Code is available at https://github.com/Yuliang-Liu/Monkey.

Jiajun Fei, Ziyu Zhu, Wenlei Liu et al. · tsinghua

Existing permutation-invariant methods can be divided into two categories according to the aggregation scope, i.e. global aggregation and local one. Although the global aggregation methods, e. g., PointNet and Deep Sets, get involved in simpler structures, their performance is poorer than the local aggregation ones like PointNet++ and Point Transformer. It remains an open problem whether there exists a global aggregation method with a simple structure, competitive performance, and even much fewer parameters. In this paper, we propose a novel global aggregation permutation-invariant network based on dual MLP dot-product, called DuMLP-Pin, which is capable of being employed to extract features for set inputs, including unordered or unstructured pixel, attribute, and point cloud data sets. We strictly prove that any permutation-invariant function implemented by DuMLP-Pin can be decomposed into two or more permutation-equivariant ones in a dot-product way as the cardinality of the given input set is greater than a threshold. We also show that the DuMLP-Pin can be viewed as Deep Sets with strong constraints under certain conditions. The performance of DuMLP-Pin is evaluated on several different tasks with diverse data sets. The experimental results demonstrate that our DuMLP-Pin achieves the best results on the two classification problems for pixel sets and attribute sets. On both the point cloud classification and the part segmentation, the accuracy of DuMLP-Pin is very close to the so-far best-performing local aggregation method with only a 1-2% difference, while the number of required parameters is significantly reduced by more than 85% in classification and 69% in segmentation, respectively. The code is publicly available on https://github.com/JaronTHU/DuMLP-Pin.

Runsheng Xu, Xin Xia, Jinlong Li et al.

Modern perception systems of autonomous vehicles are known to be sensitive to occlusions and lack the capability of long perceiving range. It has been one of the key bottlenecks that prevents Level 5 autonomy. Recent research has demonstrated that the Vehicle-to-Vehicle (V2V) cooperative perception system has great potential to revolutionize the autonomous driving industry. However, the lack of a real-world dataset hinders the progress of this field. To facilitate the development of cooperative perception, we present V2V4Real, the first large-scale real-world multi-modal dataset for V2V perception. The data is collected by two vehicles equipped with multi-modal sensors driving together through diverse scenarios. Our V2V4Real dataset covers a driving area of 410 km, comprising 20K LiDAR frames, 40K RGB frames, 240K annotated 3D bounding boxes for 5 classes, and HDMaps that cover all the driving routes. V2V4Real introduces three perception tasks, including cooperative 3D object detection, cooperative 3D object tracking, and Sim2Real domain adaptation for cooperative perception. We provide comprehensive benchmarks of recent cooperative perception algorithms on three tasks. The V2V4Real dataset can be found at https://research.seas.ucla.edu/mobility-lab/v2v4real/.

Dong Jing, Shuo Zhang, Song Chang et al.

The Light Field Raindrop Removal (LFRR) aims to restore the background areas obscured by raindrops in the Light Field (LF). Compared with single image, the LF provides more abundant information by regularly and densely sampling the scene. Since raindrops have larger disparities than the background in the LF, the majority of texture details occluded by raindrops are visible in other views. In this paper, we propose a novel LFRR network by directly utilizing the complementary pixel information of raindrop-free areas in the input raindrop LF, which consists of the re-sampling module and the refinement module. Specifically, the re-sampling module generates a new LF which is less polluted by raindrops through re-sampling position predictions and the proposed 4D interpolation. The refinement module improves the restoration of the completely occluded background areas and corrects the pixel error caused by 4D interpolation. Furthermore, we carefully build the first real scene LFRR dataset for model training and validation. Experiments demonstrate that the proposed method can effectively remove raindrops and achieves state-of-the-art performance in both background restoration and view consistency maintenance.

Zeqi Shen, Shuo Zhang, Zhuhao Zhang et al.

The Light Field (LF) deblurring task is a challenging problem as the blur images are caused by different reasons like the camera shake and the object motion. The single image deblurring method is a possible way to solve this problem. However, since it deals with each view independently and cannot effectively utilize and maintain the LF structure, the restoration effect is usually not ideal. Besides, the LF blur is more complex because the degree is affected by the views and depth. Therefore, we carefully designed a novel LF deblurring network based on the LF blur characteristics. On one hand, since the blur degree varies a lot in different views, we design a novel view adaptive spatial convolution to deblur blurred LFs, which calculates the exclusive convolution kernel for each view. On the other hand, because the blur degree also varies with the depth of the object, a depth perception view attention is designed to deblur different depth areas by selectively integrating information from different views. Besides, we introduce an angular position embedding to maintain the LF structure better, which ensures the model correctly restores the view information. Quantitative and qualitative experimental results on synthetic and real images show that the deblurring effect of our method is better than other state-of-the-art methods.

Xinyue Liu, Jianyuan Wang, Biao Leng et al.

Knowledge distillation based on student-teacher network is one of the mainstream solution paradigms for the challenging unsupervised Anomaly Detection task, utilizing the difference in representation capabilities of the teacher and student networks to implement anomaly localization. However, over-generalization of the student network to the teacher network may lead to negligible differences in representation capabilities of anomaly, thus affecting the detection effectiveness. Existing methods address the possible over-generalization by using differentiated students and teachers from the structural perspective or explicitly expanding distilled information from the content perspective, which inevitably result in an increased likelihood of underfitting of the student network and poor anomaly detection capabilities in anomaly center or edge. In this paper, we propose Dual-Modeling Decouple Distillation (DMDD) for the unsupervised anomaly detection. In DMDD, a Decouple Student-Teacher Network is proposed to decouple the initial student features into normality and abnormality features. We further introduce Dual-Modeling Distillation based on normal-anomaly image pairs, fitting normality features of anomalous image and the teacher features of the corresponding normal image, widening the distance between abnormality features and the teacher features in anomalous regions. Synthesizing these two distillation ideas, we achieve anomaly detection which focuses on both edge and center of anomaly. Finally, a Multi-perception Segmentation Network is proposed to achieve focused anomaly map fusion based on multiple attention. Experimental results on MVTec AD show that DMDD surpasses SOTA localization performance of previous knowledge distillation-based methods, reaching 98.85% on pixel-level AUC and 96.13% on PRO.

Xinyue Liu, Jianyuan Wang, Biao Leng et al.

Knowledge distillation (KD) has been widely studied in unsupervised image Anomaly Detection (AD), but its application to unsupervised multimodal AD remains underexplored. Existing KD-based methods for multimodal AD that use fused multimodal features to obtain teacher representations face challenges. Anomalies that only exist in one modality may not be effectively captured in the fused teacher features, leading to detection failures. Besides, these methods do not fully leverage the rich intra- and inter-modality information that are critical for effective anomaly detection. In this paper, we propose Tuned Reverse Distillation (TRD) based on Multi-branch design to realize Multimodal Industrial AD. By assigning independent branches to each modality, our method enables finer detection of anomalies within each modality. Furthermore, we enhance the interaction between modalities during the distillation process by designing two Crossmodal Tuners including Crossmodal Filter and Amplifier. With the idea of crossmodal mapping, the student network is allowed to better learn normal features while anomalies in all modalities are ensured to be effectively detected. Experimental verifications on multimodal AD datasets demonstrate that our method achieves state-of-the-art performance in multimodal anomaly detection and localization. Code is available at https://github.com/hito2448/TRD.

Xinyue Liu, Jianyuan Wang, Biao Leng et al.

Knowledge Distillation (KD) is a promising approach for unsupervised Anomaly Detection (AD). However, the student network's over-generalization often diminishes the crucial representation differences between teacher and student in anomalous regions, leading to detection failures. To addresses this problem, the widely accepted Reverse Distillation (RD) paradigm designs the asymmetry teacher and student, using an encoder as teacher and a decoder as student. Yet, the design of RD does not ensure that the teacher encoder effectively distinguishes between normal and abnormal features or that the student decoder generates anomaly-free features. Additionally, the absence of skip connections results in a loss of fine details during feature reconstruction. To address these issues, we propose RD with Expert, which introduces a novel Expert-Teacher-Student network for simultaneous distillation of both the teacher encoder and student decoder. The added expert network enhances the student's ability to generate normal features and optimizes the teacher's differentiation between normal and abnormal features, reducing missed detections. Additionally, Guided Information Injection is designed to filter and transfer features from teacher to student, improving detail reconstruction and minimizing false positives. Experiments on several benchmarks prove that our method outperforms existing unsupervised AD methods under RD paradigm, fully unlocking RD's potential.

Hao Wang, Shuo Zhang, Biao Leng

The computer vision community has witnessed an extensive exploration of vision transformers in the past two years. Drawing inspiration from traditional schemes, numerous works focus on introducing vision-specific inductive biases. However, the implicit modeling of permutation invariance and fully-connected interaction with individual tokens disrupts the regional context and spatial topology, further hindering higher-order modeling. This deviates from the principle of perceptual organization that emphasizes the local groups and overall topology of visual elements. Thus, we introduce the concept of hypergraph for perceptual exploration. Specifically, we propose a topology-aware vision transformer called HyperGraph Transformer (HGFormer). Firstly, we present a Center Sampling K-Nearest Neighbors (CS-KNN) algorithm for semantic guidance during hypergraph construction. Secondly, we present a topology-aware HyperGraph Attention (HGA) mechanism that integrates hypergraph topology as perceptual indications to guide the aggregation of global and unbiased information during hypergraph messaging. Using HGFormer as visual backbone, we develop an effective and unitive representation, achieving distinct and detailed scene depictions. Empirical experiments show that the proposed HGFormer achieves competitive performance compared to the recent SoTA counterparts on various visual benchmarks. Extensive ablation and visualization studies provide comprehensive explanations of our ideas and contributions.

Shuo Zhang

Saliency Prediction aims to predict the attention distribution of human eyes given an RGB image. Most of the recent state-of-the-art methods are based on deep image feature representations from traditional CNNs. However, the traditional convolution could not capture the global features of the image well due to its small kernel size. Besides, the high-level factors which closely correlate to human visual perception, e.g., objects, color, light, etc., are not considered. Inspired by these, we propose a Transformer-based method with semantic segmentation as another learning objective. More global cues of the image could be captured by Transformer. In addition, simultaneously learning the object segmentation simulates the human visual perception, which we would verify in our investigation of human gaze control in cognitive science. We build an extra decoder for the subtask and the multiple tasks share the same Transformer encoder, forcing it to learn from multiple feature spaces. We find in practice simply adding the subtask might confuse the main task learning, hence Multi-task Attention Module is proposed to deal with the feature interaction between the multiple learning targets. Our method achieves competitive performance compared to other state-of-the-art methods.

Qingyan Sun, Shuo Zhang, Song Chang et al.

Light field cameras have been proved to be powerful tools for 3D reconstruction and virtual reality applications. However, the limited resolution of light field images brings a lot of difficulties for further information display and extraction. In this paper, we introduce a novel learning-based framework to improve the spatial resolution of light fields. First, features from different dimensions are parallelly extracted and fused together in our multi-dimension fusion architecture. These features are then used to generate dynamic filters, which extract subpixel information from micro-lens images and also implicitly consider the disparity information. Finally, more high-frequency details learned in the residual branch are added to the upsampled images and the final super-resolved light fields are obtained. Experimental results show that the proposed method uses fewer parameters but achieves better performances than other state-of-the-art methods in various kinds of datasets. Our reconstructed images also show sharp details and distinct lines in both sub-aperture images and epipolar plane images.

Hongwei Xie, Shuo Zhang, Huanghao Ding et al.

The inherent heavy computation of deep neural networks prevents their widespread applications. A widely used method for accelerating model inference is quantization, by replacing the input operands of a network using fixed-point values. Then the majority of computation costs focus on the integer matrix multiplication accumulation. In fact, high-bit accumulator leads to partially wasted computation and low-bit one typically suffers from numerical overflow. To address this problem, we propose an overflow aware quantization method by designing trainable adaptive fixed-point representation, to optimize the number of bits for each input tensor while prohibiting numeric overflow during the computation. With the proposed method, we are able to fully utilize the computing power to minimize the quantization loss and obtain optimized inference performance. To verify the effectiveness of our method, we conduct image classification, object detection, and semantic segmentation tasks on ImageNet, Pascal VOC, and COCO datasets, respectively. Experimental results demonstrate that the proposed method can achieve comparable performance with state-of-the-art quantization methods while accelerating the inference process by about 2 times.

Shuo Zhang, Krisztian Balog

Tables are an extremely powerful visual and interactive tool for structuring and manipulating data, making spreadsheet programs one of the most popular computer applications. In this paper we introduce and address the task of recommending related tables: given an input table, identifying and returning a ranked list of relevant tables. One of the many possible application scenarios for this task is to provide users of a spreadsheet program proactively with recommendations for related structured content on the Web. At its core, the related table recommendation task boils down to computing the similarity between a pair of tables. We develop a theoretically sound framework for performing table matching. Our approach hinges on the idea of representing table elements in multiple semantic spaces, and then combining element-level similarities using a discriminative learning model. Using a purpose-built test collection from Wikipedia tables, we demonstrate that the proposed approach delivers state-of-the-art performance.

Yao Sun, Yang Deng, Yue Xu et al.

Magnetic Resonance Imaging (MRI) is widely used in the pathological and functional studies of the brain, such as epilepsy, tumor diagnosis, etc. Automated accurate brain tissue segmentation like cerebro-spinal fluid (CSF), gray matter (GM), white matter (WM) is the basis of these studies and many researchers are seeking it to the best. Based on the truth that multi-channel segmentation network with its own ground truth achieves up to average dice ratio 0.98, we propose a novel method that we add a fourth channel with the ground truth of the most similar image's obtained by CBIR from the database. The results show that the method improves the segmentation performance, as measured by average dice ratio, by approximately 0.01 in the MRBrainS18 database. In addition, our method is concise and robust, which can be used to any network architecture that needs not be modified a lot.

Yang Deng, Yao Sun, Yongpei Zhu et al.

Accurate segmentation of MR brain tissue is a crucial step for diagnosis, surgical planning, and treatment of brain abnormalities. Automatic and reliable segmenta-tion methods are required to assist doctor. Over the last few years, deep learning especially deep convolutional neural networks (CNNs) have emerged as one of the most prominent approaches for image recognition problems in various do-mains. But the improvement of deep networks always needs inspiration, which is rare for the ordinary. Until now,there have been reasonable MR brain tissue segmentation methods,all of which can achieve promising performance. These different methods have their own characteristic and are distinctive for data sets. In other words, different models performance vary widely on the same data sets and each model has what it is skilled in. It is on the basis of this, we propose a judgement to distinguish data sets that different models are good at. With our method, the segmentation accuracy can be improved easily based on the existing models, neither without increasing training data nor improving the network. We validate our method on the widely used IBSR 18 dataset and obtain average dice ratio of 88.06%,while it is 85.82% and 86.92% when only using separate one model respectively.