arXiv subjects

Yanhua Yang

Publications and source records attributed to Yanhua Yang.

6 recordsLinked to original sources

arXiv2026

Your AI-Generated Image Detector Can Secretly Achieve SOTA Accuracy, If Calibrated

Despite being trained on balanced datasets, existing AI-generated image detectors often exhibit systematic bias at test time, frequently misclassifying fake images as real. We hypothesize that this behavior stems from distributional shift in fake samples and implicit priors learned during training. Specifically, models tend to overfit to superficial artifacts that do not generalize well across different generation methods, leading to a misaligned decision threshold when faced with test-time distribution shift. To address this, we propose a theoretically grounded post-hoc calibration framework based on Bayesian decision theory. In particular, we introduce a learnable scalar correction to the model's logits, optimized on a small validation set from the target distribution while keeping the backbone frozen. This parametric adjustment compensates for distributional shift in model output, realigning the decision boundary even without requiring ground-truth labels. Experiments on challenging benchmarks show that our approach significantly improves robustness without retraining, offering a lightweight and principled solution for reliable and adaptive AI-generated image detection in the open world. Code is available at https://github.com/muliyangm/AIGI-Det-Calib.

cs.CV↗

arXiv2022

Progressive Domain-Independent Feature Decomposition Network for Zero-Shot Sketch-Based Image Retrieval

Zero-shot sketch-based image retrieval (ZS-SBIR) is a specific cross-modal retrieval task for searching natural images given free-hand sketches under the zero-shot scenario. Most existing methods solve this problem by simultaneously projecting visual features and semantic supervision into a low-dimensional common space for efficient retrieval. However, such low-dimensional projection destroys the completeness of semantic knowledge in original semantic space, so that it is unable to transfer useful knowledge well when learning semantic from different modalities. Moreover, the domain information and semantic information are entangled in visual features, which is not conducive for cross-modal matching since it will hinder the reduction of domain gap between sketch and image. In this paper, we propose a Progressive Domain-independent Feature Decomposition (PDFD) network for ZS-SBIR. Specifically, with the supervision of original semantic knowledge, PDFD decomposes visual features into domain features and semantic ones, and then the semantic features are projected into common space as retrieval features for ZS-SBIR. The progressive projection strategy maintains strong semantic supervision. Besides, to guarantee the retrieval features to capture clean and complete semantic information, the cross-reconstruction loss is introduced to encourage that any combinations of retrieval features and domain features can reconstruct the visual features. Extensive experiments demonstrate the superiority of our PDFD over state-of-the-art competitors.

cs.CV↗

arXiv2021

Understanding and Improving Early Stopping for Learning with Noisy Labels

The memorization effect of deep neural network (DNN) plays a pivotal role in many state-of-the-art label-noise learning methods. To exploit this property, the early stopping trick, which stops the optimization at the early stage of training, is usually adopted. Current methods generally decide the early stopping point by considering a DNN as a whole. However, a DNN can be considered as a composition of a series of layers, and we find that the latter layers in a DNN are much more sensitive to label noise, while their former counterparts are quite robust. Therefore, selecting a stopping point for the whole network may make different DNN layers antagonistically affected each other, thus degrading the final performance. In this paper, we propose to separate a DNN into different parts and progressively train them to address this problem. Instead of the early stopping, which trains a whole DNN all at once, we initially train former DNN layers by optimizing the DNN with a relatively large number of epochs. During training, we progressively train the latter DNN layers by using a smaller number of epochs with the preceding layers fixed to counteract the impact of noisy labels. We term the proposed method as progressive early stopping (PES). Despite its simplicity, compared with the early stopping, PES can help to obtain more promising and stable results. Furthermore, by combining PES with existing approaches on noisy label training, we achieve state-of-the-art performance on image classification benchmarks.

cs.LG↗

arXiv2019

On-node lattices construction using $\textit{partial}$ Gauss-Hermite quadrature for the lattice Boltzmann method

A concise theoretical framework, the $\textit{partial}$ Gauss-Hermite quadrature (pGHQ), is established for constructing on-node lattices of the lattice Boltzmann (LB) method under a Cartesian coordinate system. Comparing with existing approaches, the pGHQ scheme has the following advantages: $\textbf{a).}$ extremely concise algorithm, $\textbf{b).}$ unifying the constructing procedure of symmetric and asymmetric on-node lattices, $\textbf{c).}$ covering full-range quadrature degree of a given discrete velocity set. We employ it to search the local optimal and asymmetric lattices for $\left\{ {n = 3,4,5,6,7} \right\}$ moment degree equilibrium distribution discretization on range $\left[ { - 10,10} \right]$. The search reveals a surprising abundance of available lattices. Through a brief analysis, the discrete velocity set shows a significant influence on the positivity of equilibrium distributions, which is considered as one major impact to the numerical stability of the LB method. Hence the results of the pGHQ scheme lay a foundation for further investigations on improving the numerical stability of the LB method by modifying the discrete velocity set. It also worths noting that pGHQ can be extended into the entropic LB model though it was proposed for the Hermite polynomial expansion LB theory.

physics.comp-ph↗

arXiv2019

Derivation of lattice Boltzmann equation via analytical characteristic integral

A lattice Boltzmann (LB) theory, analytical characteristic integral (ACI) LB theory, is proposed in this paper. ACI LB theory takes Bhatnagar-Gross-Krook (BGK) Boltzmann equation as the exact kinetic equation behind Navier-Stokes continuum and momentum equations and constructs LB equation by rigorously integrating BGK-Boltzmann equation along characteristics. It's a general theory, supporting most existed LB equations including the standard lattice BGK (LBGK) equation inherited from lattice-gas automata, whose theoretical foundation had been questioned. ACI LB theory also indicates that the characteristic parameter of LB equation is collision number, depicting the particle-interacting intensity in the time span of LB equation, instead of traditionally assumed relaxation time, and the over relaxation time problem is merely a manifestation of temporal evolution of equilibrium distribution along characteristics under high collision number, irrelevant to particle kinetics. In ACI LB theory, the temporal evolution of equilibrium distribution along characteristics is the determinant of LB method accuracy and we numerically prove it.

physics.flu-dyn↗

arXiv2017

On the derivations of lattice Boltzmann evolution equation

A comparative analysis on the popular schemes for evaluating evolution equation in lattice Boltzmann method (LBM) is presented in this paper. It includes two classical characteristic-line schemes, Boesh-Karlin and He-Luo scheme, and a author-proposed scheme, Taylor-expansion scheme, originating from the extension of He-Luo scheme. We detailly discuss the mathematical mechanism and the equilibrium distribution evolution behind them. By analyzing the conflict between prediction and derivation, we address the preconditions for these schemes. At the end, we conclude their pros and cons and suggest scheme's applicable scene based on their derivation procedure and further development capacity.

physics.comp-ph↗