Reconstructing large-scale dynamic scenes from visual observations is a fundamental challenge in computer vision, with critical implications for robotics and autonomous systems. While recent differentiable rendering methods such as Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) have achieved impressive photorealistic reconstruction, they suffer from scalability limitations and require annotations to decouple actor motion. Existing self-supervised methods attempt to eliminate explicit annotations by leveraging motion cues and geometric priors, yet they remain constrained by per-scene optimization and sensitivity to hyperparameter tuning. In this paper, we introduce Flux4D, a simple and scalable framework for 4D reconstruction of large-scale dynamic scenes. Flux4D directly predicts 3D Gaussians and their motion dynamics to reconstruct sensor observations in a fully unsupervised manner. By adopting only photometric losses and enforcing an "as static as possible" regularization, Flux4D learns to decompose dynamic elements directly from raw data without requiring pre-trained supervised models or foundational priors simply by training across many scenes. Our approach enables efficient reconstruction of dynamic scenes within seconds, scales effectively to large datasets, and generalizes well to unseen environments, including rare and unknown objects. Experiments on outdoor driving datasets show Flux4D significantly outperforms existing methods in scalability, generalization, and reconstruction quality.
PatchMVSNet: Patch-wise Unsupervised Multi-View Stereo for Weakly-Textured Surface Reconstruction
Wuhan University（武汉大学）
基于学习的多视图立体视觉（MVS）在流行的数据集上获得了良好的重建效果。然而，监督学习方法需要用于训练的基础事实，这是难以收集的，特别是对于大规模数据集。虽然目前已经提出了无监督学习方法并取得了令人满意的结果，但这些方法仍然无法在具有挑战性的场景中重建完整的结果，例如弱纹理表面，因为这些方法主要依赖于受到各种光照的像素级光度一致性。为了减少这些具有挑战性的场景中的匹配模糊性，本文提出了鲁棒的损失函数，利用多视图图像下的约束：1）逐块光度一致性损失，这扩大了多视图相似性测量中的特征的感受野，2）鲁棒的两视图几何一致性，这包括最小遮挡的跨视图深度一致性检查。我们的无监督策略可以用任意的深度估计框架来实现，并且可以用任意的大规模MVS数据集来训练。实验结果表明，该方法能有效降低匹配二义性，特别是提高弱纹理重建的完整性。此外，我们的方法在流行的基准测试中达到了最先进方法的性能，如DTU，Tanks and Temples和ETH3D。代码将很快发布。
Multi-Angle Point Cloud-VAE: Unsupervised Feature Learning for 3D Point Clouds from Multiple Angles by Joint Self-Reconstruction and Half-to-Half Prediction
Tsinghua University（清华大学）；University of Maryland, College Park（马里兰大学帕克分校）；Beijing National Research Center for Information Science and Technology (BNRist)（北京信息科学与技术国家研究中心（BNRist））
Unsupervised 3D Reconstruction from a Single Image via Adversarial Learning
NYU Multimedia and Visual Computing Lab（纽约大学多媒体与视觉计算实验室）；Courant Institute of Mathematical Science（库朗数学科学研究所）；NYU Tandon School of Engineering（纽约大学坦登工程学院）；NYU Abu Dhabi（纽约大学阿布扎比分校）
Model-based Reconstruction with Learning: From Unsupervised to Supervised and Beyond
Michigan State University（密歇根州立大学）；University of Michigan - Shanghai Jiao Tong University Joint Institute（密歇根大学-上海交通大学联合学院）；Shanghai Jiao Tong University（上海交通大学）
Multi-feature Reconstruction Network using Crossed-mask Restoration for Unsupervised Industrial Anomaly Detection
School of Information and Communication Engineering, Zhongyuan University of Technology（中原工学院信息与通信工程学院）；College of Automation Engineering, Nanjing University of Aeronautics and Astronautics（南京航空航天大学自动化学院）；Applied Technology Research Institute, Henan Institute of Economics and Trade（河南经贸职业学院应用技术研究院）
Enhancing Low-dose CT Image Reconstruction by Integrating Supervised and Unsupervised Learning
University of Michigan Shanghai Jiao Tong University Joint Institute（密歇根大学上海交通大学联合学院）；Shanghai Jiao Tong University（上海交通大学）；Amazon Inc.（亚马逊公司）；Michigan State University（密歇根州立大学）
DULDA: Dual-domain Unsupervised Learned Descent Algorithm for PET image reconstruction
Zhejiang University（浙江大学）；Massachusetts General Hospital（麻省总医院）；Harvard Medical School（哈佛医学院）；University of Florida（佛罗里达大学）；Massachusetts General Brigham（麻省总医院布莱根医疗系统）
Unsupervised Spatial-spectral Hyperspectral Image Reconstruction and Clustering with Diffusion Geometry
City University of Hong Kong（香港城市大学）；Hong Kong Centre for Cerebro-Cardiovascular Health Engineering（香港脑心血管健康工程中心）；Tufts University（塔夫茨大学）；Wake Forest University（维克森林大学）
DPF-Net: Combining Explicit Shape Priors in Deformable Primitive Field for Unsupervised Structural Reconstruction of 3D Objects
National Engineering Laboratory for Brain-inspired Intelligence Technology and Application（脑启发智能技术与应用国家工程实验室）；University of Science and Technology of China（中国科学技术大学）
MADGAN: unsupervised Medical Anomaly Detection GAN using multiple adjacent brain MRI slice reconstruction
LPIXEL Inc.（LPIXEL公司）；University of Cambridge（剑桥大学）；National Institute of Informatics（国立信息学研究所）；National Center for Global Health and Medicine（国立国际医疗研究中心）；Juntendo University（顺天堂大学）；ELTE Eötvös Loránd University（厄特沃什·罗兰大学）；The University of Tokyo（东京大学）
Unsupervised Low-dose CT Reconstruction with One-way Conditional Normalizing Flows
School of Mathematical Sciences, Capital Normal University（首都师范大学数学学院）；Centre for Mathematical Imaging Techniques, University of Liverpool（利物浦大学数学成像技术中心）；Department of Mathematics and Statistics, University of Strathclyde（斯特拉思克莱德大学数学与统计学系）
School of Biomedical Engineering, Shanghai Jiao Tong University（上海交通大学生物医学工程学院）；School of Information Science and Technology, ShanghaiTech University（上海交通大学信息科学与技术学院）；School of Biomedical Engineering and the National Engineering Research Center of Advanced Magnetic Resonance Technologies for Diagnosis and Therapy (NERCAMRT), Shanghai Jiao Tong University（上海交通大学生物医学工程学院和先进磁共振诊断与治疗国家工程研究中心）
SwinAD: Multi-stage feature reconstruction for unsupervised industrial anomaly detection
Institute of Artificial Intelligence, University of Engineering and Technology, Vietnam National University（越南国立大学工程技术大学人工智能研究所）；Human-Machine Interaction Laboratory, University of Engineering and Technology, Vietnam National University（越南国立大学工程技术大学人机交互实验室）；Applied AI Lab, Phenikaa School of Computing, Phenikaa University（费尼卡大学计算学院应用人工智能实验室）
Unsupervised denoising for sparse multi-spectral computed tomography
University of Oulu（奥卢大学）；University College London（伦敦大学学院）
具有光子计数探测器（PCD）的多能量计算机断层扫描（CT）能够实现光谱成像，因为PCD可以将入射光子分配给特定的能量通道。然而，具有许多光谱通道的PCD大幅增加了CT重建的计算复杂度，并且定制重建算法需要根据不同的噪声统计进行微调。\rev{Especially if many projections are taken, a large amount of data has to be collected and stored. Sparse view CT is one solution for data reduction. However, these issues are especially exacerbated when sparse imaging scenarios are encountered due to a significant reduction in photon counts.}在这项工作中，我们研究了基于学习的改进，以获得高质量的重建从稀疏测量的64通道PCD-CT的挑战性任务的适用性。特别是，为了克服训练过程中丢失的参考数据，我们提出了一种无监督的去噪和伪影去除方法，通过在重建中利用不同的滤波器函数以及光谱通道与核范数的显式耦合。性能评估模拟合成数据和公开提供的实验多光谱成像通过计算机断层扫描（MUSIC）数据集。我们将我们的无监督方法的质量…
Unsupervised Learning on a DIET: Datum IndEx as Target Free of Self-Supervision, Reconstruction, Projector Head
Meta AI；Cranberry-Lemon University（蔓越莓柠檬大学）；University of the Witwatersrand（威特沃特斯兰德大学）
如果我们希望学习廉价、可靠和可转移的模型，那么代价高昂、嘈杂和过度专业化的标签将被搁置一边，转而支持无监督学习。为此，谱嵌入、自监督学习或生成式建模提供了有竞争力的解决方案。然而，这些方法带来了许多挑战，\textit{e.g.}估计测地线距离，指定投影仪架构和抗崩溃损失，或指定解码器架构和重建损失。相比之下，我们引入了一个简单的可解释的替代方案-创造的\textbf{DIET} -从未标记的数据中学习表示，没有这些挑战。\textbf{DIET}非常简单：采用自己喜欢的分类设置，并使用\textbf{D} atum \textbf{I} nd \textbf{E} x作为其\textbf{T}目标类，\textit{i.e. each sample is its own class}，无需进一步更改。\textbf{DIET}不需要解码器/投影器网络，不基于正对也不基于重构，不引入超参数，并且可以跨数据集和架构进行开箱即用。尽管\textbf{DIET}很简单，但学习到的表示具有高质量，并且通常与最先进的\textit{e.g.}相当，在DIET的学习表示之上使用线性分类器，在CIFAR 100上使用Resnet 101达到$71.4\%$，在TinyImagenet上使用Resnext 50达到$52.5\%$。
