Author name cluster

Kai Xiao

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

7 papers

2 author rows

EAAI Journal 2025 Journal Article

An Efficient Point Network for Light Detection and Ranging point cloud perception in large-scale scene

Kai Xiao
Jialin Gui
Yuanxi Peng
Teng Li

With the ongoing advancement and wide application of Light Detection and Ranging (LiDAR) technology, the efficient deployment of deep learning models in various offline edge scenes, such as autonomous vehicles and drones, has become an important research hotspot in industry and academia. How to achieve lightweight and efficient network model (light and fast) on edge devices with limited resources is the key point. Model quantization methods, benefiting from lightweight parameters, have become one of the most promising approaches for model compression and acceleration. However, due to the unstructured characteristic of the point cloud, quantization may lead to aggregation-induced feature homogenization and structural scale distortion, resulting in performance bottlenecks for quantization models. Moreover, unstructured point cloud is difficult to adapt to existing image processing quantification methods and require a new network architecture. To this end, we constructed a specialized point cloud quantization network and designed an effective feature extraction architecture, proposing an Efficient Point Network (EPoNET). Specifically, to address the issues of feature homogenization and limited feature discriminability after quantization, we designed the dual refining kernel and the local feature reshaping module to refine and expand the features. To tackle the class imbalance within the dataset, we designed the weighted loss function to enhance the effectiveness of the network training and alleviate the interference caused by local optima during training. Finally, we evaluated the proposed EPoNET using the large-scale, real-world multispectral LiDAR point cloud dataset. This dataset covers about 25 square kilometers and includes 8. 52 million points of different land features like roads, trees, and buildings. Experimental results demonstrate that EPoNET achieves an overall accuracy surpassing 90%, with an average accuracy of 91. 31% across the testing areas. Compared with the lightweight benchmark method PointNet++, EPoNET brings 5. 7x inference acceleration and 58. 4% usage savings, achieving a balance between model inference accuracy, speed, and computational consumption.

Details DOI

EAAI Journal 2025 Journal Article

Multi-stream feature aggregation network with multi-scale supervision for single image dehazing

Junjiang Wu
Haibo Tao
Kai Xiao
Jun Chu
Lu Leng

Single image dehazing is a challenge, as it requires to eliminate the haze while preserving image quality. Most existing models use encoder–decoder or single-scale structures. Owing to flaws of architectural design, these models fail to capture spatial details and semantic contexts in a complementary manner, resulting in inaccurately detailed recovery and suboptimal dehazing performance. To address these issues, we propose a novel framework known as the Multi-stream Feature Aggregation Network (MSFANet), which leverages and excavates the information from different scales of input images and fuses different levels of features. In addition, we design a Self-adjustable Complementary Features Selection Module (SCFSM) that efficiently selects and aggregates the features with different scales from multiple streams, enabling the full exchange of information between different levels of features within the network. Moreover, during the training stage, we introduce a Fast Fourier Transform (FFT) loss combined with pixel loss to simultaneously supervise the reconstruction process in the spatial and frequency domains with a multi-scale supervision strategy. Extensive experiments on homogeneous and nonhomogeneous datasets show that the proposed MSFANet achieves state-of-the-art dehazing performance. Specifically, our method dramatically boosts the Peak Signal-to-Noise Ratio (PSNR) metric to 42. 47 dB on the commonly used Synthetic Objective Testing Set (SOTS) indoor dataset.

Details DOI

EAAI Journal 2024 Journal Article

A spatial–temporal contexts network for object tracking

Kai Huang
Kai Xiao
Jun Chu
Lu Leng
Xingbo Dong

Although there have been significant advancements and developments in visual object tracking in recent years, most trackers have failed to adapt to the deterioration of object appearance in complex scenes. Typically, they utilize only spatial information or simple temporal networks. The fusion of spatial and temporal contexts among consecutive frames can hypothetically capture historical information to boost tracking performance but inevitably pollutes the model with noisy samples. To this end, we proposed a novel end-to-end ConvLSTM-based tracking framework called STCTrack, which uses spatial and temporal information from each frame and adapts to noisy samples. Specifically, a multilayer residual ConvLSTM-based spatial–temporal context network (STCN) was proposed in STCTrack to retain the target’s past information and consequently guide the tracker to focus on the most informative regions of the current frame. Furthermore, a multi-similarity map fusion model was proposed to calculate the pixel-level similarity map, allowing STCTrack to adaptively retrieve historical target information from different times and be resilient to partial occlusions and nonrigid deformations. Extensive empirical studies were conducted on benchmarks, including OTB2015, GOT-10K, TrackingNet, LaSOT, UAV123, and VOT2018. The empirical results suggest that the proposed STCTrack achieves state-of-the-art performance compared with existing schemes. The code and models used in this study are publicly available at www. github. com/Kevoen/STCTrack to encourage further research on this topic.

Details DOI

NeurIPS Conference 2022 Conference Paper

3DB: A Framework for Debugging Computer Vision Models

Guillaume Leclerc
Hadi Salman
Andrew Ilyas
Sai Vemprala
Logan Engstrom
Vibhav Vineet
Kai Xiao
Pengchuan Zhang

We introduce 3DB: an extendable, unified framework for testing and debugging vision models using photorealistic simulation. We demonstrate, through a wide range of use cases, that 3DB allows users to discover vulnerabilities in computer vision systems and gain insights into how models make decisions. 3DB captures and generalizes many robustness analyses from prior work, and enables one to study their interplay. Finally, we find that the insights generated by the system transfer to the physical world. 3DB will be released as a library alongside a set of examples and documentation. We attach 3DB to the submission.

PDF Details

ICLR Conference 2020 Conference Paper

A Framework for robustness Certification of Smoothed Classifiers using F-Divergences

Krishnamurthy Dvijotham
Jamie Hayes
Borja Balle
J. Zico Kolter
Chongli Qin
András György 0001
Kai Xiao
Sven Gowal

Formal verification techniques that compute provable guarantees on properties of machine learning models, like robustness to norm-bounded adversarial perturbations, have yielded impressive results. Although most techniques developed so far require knowledge of the architecture of the machine learning model and remain hard to scale to complex prediction pipelines, the method of randomized smoothing has been shown to overcome many of these obstacles. By requiring only black-box access to the underlying model, randomized smoothing scales to large architectures and is agnostic to the internals of the network. However, past work on randomized smoothing has focused on restricted classes of smoothing measures or perturbations (like Gaussian or discrete) and has only been able to prove robustness with respect to simple norm bounds. In this paper we introduce a general framework for proving robustness properties of smoothed machine learning models in the black-box setting. Specifically, we extend randomized smoothing procedures to handle arbitrary smoothing measures and prove robustness of the smoothed classifier by using f-divergences. Our methodology improves upon the state of the art in terms of computation time or certified robustness on several image classification tasks and an audio classification task, with respect to several classes of adversarial perturbations.

Details

ICLR Conference 2020 Conference Paper

Toward Evaluating Robustness of Deep Reinforcement Learning with Continuous Control

Tsui-Wei Weng
Krishnamurthy Dvijotham
Jonathan Uesato
Kai Xiao
Sven Gowal
Robert Stanforth
Pushmeet Kohli

Deep reinforcement learning has achieved great success in many previously difficult reinforcement learning tasks, yet recent studies show that deep RL agents are also unavoidably susceptible to adversarial perturbations, similar to deep neural networks in classification tasks. Prior works mostly focus on model-free adversarial attacks and agents with discrete actions. In this work, we study the problem of continuous control agents in deep RL with adversarial attacks and propose the first two-step algorithm based on learned model dynamics. Extensive experiments on various MuJoCo domains (Cartpole, Fish, Walker, Humanoid) demonstrate that our proposed framework is much more effective and efficient than model-free based attacks baselines in degrading agent performance as well as driving agents to unsafe states.

Details

IS Journal 2016 Journal Article

Brain MRI Tumor Segmentation with 3D Intracranial Structure Deformation Features

Shang-Ling Jui
Shichen Zhang
Weilun Xiong
Fangxiaoqi Yu
Mingjian Fu
Dongmei Wang
Aboul Ella Hassanien
Kai Xiao

Extraction of relevant features is of significant importance for brain tumor segmentation systems. To improve brain tumor segmentation accuracy, the authors present an improved feature extraction component that takes advantage of the correlation between intracranial structure deformation and the compression resulting from brain tumor growth. Using 3D nonrigid registration and deformation modeling techniques, the component measures lateral ventricular (LaV) deformation in volumetric magnetic resonance images. By verifying the location of the extracted LaV deformation feature data and applying the features on brain tumor segmentation with widely used classification algorithms, the authors evaluate the proposed component qualitatively and quantitatively with promising results on 11 datasets comprising real and simulated patient images.

Details DOI