Search

In recent years, anchor and hash-based multi-view clustering methods have gained attention for their efficiency and simplicity in handling large-scale data. However, existing methods often overlook the interactions among multi-view data and higher-order cooperative relationships during projection, negatively impacting the quality of hash representation in low-dimensional spaces, clustering performance, and sensitivity to noise. To address this issue, we propose a novel approach named Tensor-Interacted Projection and Cooperative Hashing for Multi-View Clustering(TPCH). TPCH stacks multiple projection matrices into a tensor, taking into account the synergies and communications during the projection process. By capturing higher-order multi-view information through dual projection and Hamming space, TPCH employs an enhanced tensor nuclear norm to learn more compact and distinguishable hash representations, promoting communication within and between views. Experimental results demonstrate that this refined method significantly outperforms state-of-the-art methods in clustering on five large-scale multi-view datasets. Moreover, in terms of CPU time, TPCH achieves substantial acceleration compared to the most advanced current methods. The code is available at \textcolor{red}{\url{https://github.com/jankin-wang/TPCH}}.

A significant challenge facing current optical flow and stereo methods is the difficulty in generalizing them well to the real world. This is mainly due to the high costs required to produce datasets, and the limitations of existing self-supervised methods on fuzzy results and complex model training problems. To address the above challenges, we propose a unified self-supervised generalization framework for optical flow and stereo tasks: Self-Assessed Generation (SAG). Unlike previous self-supervised methods, SAG is data-driven, using advanced reconstruction techniques to construct a reconstruction field from RGB images and generate datasets based on it. Afterward, we quantified the confidence level of the generated results from multiple perspectives, such as reconstruction field distribution, geometric consistency, and structural similarity, to eliminate inevitable defects in the generation process. We also designed a 3D flight foreground automatic rendering pipeline in SAG to encourage the network to learn occlusion and motion foreground. Experimentally, because SAG does not involve changes to methods or loss functions, it can directly self-supervised train the state-of-the-art deep networks, greatly improving the generalization performance of self-supervised methods on current mainstream optical flow and stereo-matching datasets. Compared to previous training modes, SAG is more generalized, cost-effective, and accurate.

Perceiving and understanding 3D motion is a core technology in fields such as autonomous driving, robots, and motion prediction. This paper proposes a 3D motion perception method called ScaleFlow++ that is easy to generalize. With just a pair of RGB images, ScaleFlow++ can robustly estimate optical flow and motion-in-depth (MID). Most existing methods directly regress MID from two RGB frames or optical flow, resulting in inaccurate and unstable results. Our key insight is cross-scale matching, which extracts deep motion clues by matching objects in pairs of images at different scales. Unlike previous methods, ScaleFlow++ integrates optical flow and MID estimation into a unified architecture, estimating optical flow and MID end-to-end based on feature matching. Moreover, we also proposed modules such as global initialization network, global iterative optimizer, and hybrid training pipeline to integrate global motion information, reduce the number of iterations, and prevent overfitting during training. On KITTI, ScaleFlow++ achieved the best monocular scene flow estimation performance, reducing SF-all from 6.21 to 5.79. The evaluation of MID even surpasses RGBD-based methods. In addition, ScaleFlow++ has achieved stunning zero-shot generalization performance in both rigid and nonrigid scenes. Code is available at \url{https://github.com/HanLingsgjk/CSCV}., Comment: This is a product uploaded incorrectly. I originally intended to use it to replace ScaleRAFT (arXiv:2407.09797), but I made a mistake in the operation

Perceiving and understanding 3D motion is a core technology in fields such as autonomous driving, robots, and motion prediction. This paper proposes a 3D motion perception method called ScaleFlow++ that is easy to generalize. With just a pair of RGB images, ScaleFlow++ can robustly estimate optical flow and motion-in-depth (MID). Most existing methods directly regress MID from two RGB frames or optical flow, resulting in inaccurate and unstable results. Our key insight is cross-scale matching, which extracts deep motion clues by matching objects in pairs of images at different scales. Unlike previous methods, ScaleFlow++ integrates optical flow and MID estimation into a unified architecture, estimating optical flow and MID end-to-end based on feature matching. Moreover, we also proposed modules such as global initialization network, global iterative optimizer, and hybrid training pipeline to integrate global motion information, reduce the number of iterations, and prevent overfitting during training. On KITTI, ScaleFlow++ achieved the best monocular scene flow estimation performance, reducing SF-all from 6.21 to 5.79. The evaluation of MID even surpasses RGBD-based methods. In addition, ScaleFlow++ has achieved stunning zero-shot generalization performance in both rigid and nonrigid scenes. Code is available at \url{https://github.com/HanLingsgjk/CSCV}., Comment: 14 pages; Previously this version appeared as arXiv:2409.12202 which was submitted as a new work by accident

A significant challenge facing current optical flow methods is the difficulty in generalizing them well to the real world. This is mainly due to the high cost of hand-crafted datasets, and existing self-supervised methods are limited by indirect loss and occlusions, resulting in fuzzy outcomes. To address this challenge, we introduce a novel optical flow training framework: automatic data factory (ADF). ADF only requires RGB images as input to effectively train the optical flow network on the target data domain. Specifically, we use advanced Nerf technology to reconstruct scenes from photo groups collected by a monocular camera, and then calculate optical flow labels between camera pose pairs based on the rendering results. To eliminate erroneous labels caused by defects in the scene reconstructed by Nerf, we screened the generated labels from multiple aspects, such as optical flow matching accuracy, radiation field confidence, and depth consistency. The filtered labels can be directly used for network supervision. Experimentally, the generalization ability of ADF on KITTI surpasses existing self-supervised optical flow and monocular scene flow algorithms. In addition, ADF achieves impressive results in real-world zero-point generalization evaluations and surpasses most supervised methods., Comment: 8 pages

The Kovacs effect is a remarkable feature of the ageing dynamics of glass forming liquids near the glass transition temperature. It consists in a non-monotonous evolution of the volume/enthalpy after a succession of two abrupt temperature changes: first from a high initial temperature $T_i$ to a much lower annealing temperature $T_a$ followed by a smaller second jump back to a slightly higher final temperature $T_f$. The second change is performed when the instantaneous value of the volume/enthalpy coincides with the equilibrium one at the final temperature. While this protocol might be expected to yield equilibrium dynamics right after the second temperature change, one observes the so-called Kovacs hump in glassy systems. In this paper we apply such thermal protocol to the Distinguishable Particles Lattice Model (DPLM) for a wide range of fragility of the system. We study the Kovacs hump based on energy relaxation and all main experimental features are captured. Results are compared to general predictions based on a master equation approach in the linear response limit. We trace the origin of the Kovacs hump to the non-equilibrium nature of the probability distribution of particle interaction energies after the annealing and find that its difference with respect to the final equilibrium distribution is non-vanishing with two isolated zeros. This allows Kovacs' condition of equilibrium total energy to be met out-of-equilibrium, thus representing the memory content of the system. Furthermore, the hump is taller and occurs at a larger overlap with the system initial configuration for more fragile systems. The dynamics of a structural temperature for the mobile regions strongly depends on the glass fragility while for the immobile ones only a weak dependence is found., Comment: 9 pages, 9 figures: Added fragility study and encoding of the memory in the particles interaction energy distribution; added doi

We perform kinetic Monte Carlo simulations of a distinguishable-particle lattice model of structural glasses with random particle interactions. By varying the interaction distribution and the average particle hopping energy barrier, we obtain an extraordinarily wide range of kinetic fragility. A stretching exponent, characterizing structural relaxation, is found to decrease with the kinetic fragility in agreement with experiments. The most fragile glasses are those exhibiting low hopping barriers and, more importantly, dramatic drops of entropies upon cooling toward the glass transition temperatures. The entropy drops reduce possible kinetic pathways and lead to dramatic slowdowns in the dynamics. In addition, the kinetic fragility is shown to correlate with a thermodynamic fragility., Comment: Main text: 6 pages with 4 figures Supplementary: 11 pages with 18 figures

We report an analytical evaluation of the mean-squared displacement (MSD) of the particles in glasses based on their coarse grained trajectories. The calculation is conducted by means of a local random configuration-tree theory that was recently proposed by one of us [C.-H. Lam, J. Stat. Mech. \textbf{2018}, 023301 (2018)]. Results are compared with the numerical simulations of a lattice glass model, and good quantitative agreement has been obtained over a wide range of temperatures in the entire region of time with virtually no free parameters. To the best of our knowledge, the calculation is the first in its kind., Comment: 11 pages, 2 figures

We propose an interacting lattice gas model of structural glass characterized by particle distinguishability and site-particle-dependent random nearest-neighboring particle interactions. This incorporates disorder quenched in the configuration space rather than in the physical space. The model exhibits non-trivial energetics while still admitting exact equilibrium states directly constructible at arbitrary temperature and density. The dynamics is defined by activated hopping following standard kinetic Monte Carlo approach without explicit facilitation rule. Kinetic simulations show emergent dynamic facilitation behaviors in the glassy phase in which motions of individual voids are significant only when accelerated by other voids nearby. This provides a microscopic justification for the dynamic facilitation picture of structural glass.