arXiv subjects

Xixi Liu

Publications and source records attributed to Xixi Liu.

8 recordsLinked to original sources

arXiv2026

What Matters is the Prompt: Prompt Sensitivity and Prompt Generation in Foundation Models for Lung Nodule Segmentation

Lung nodule segmentation in computed tomography is essential for extracting clinically relevant information for lung cancer assessment and treatment planning. Foundation models have shown notable segmentation capabilities, but state-of-the-art approaches often depend on input prompts, such as points or boxes, making their performance sensitive to prompt quality and placement. Understanding the limitations and constraints of prompt-based foundation models is therefore essential for designing reliable medical image segmentation solutions. In this work, we investigate how prompt quality affects foundation models performance for lung nodule segmentation. We further propose a synthetic prompt-generation model to test if the dependence on manually provided prompts can be mitigated by generating synthetic prompts that can also improve segmentation performance. Perturbation experiments show that bounding box prompts generally outperform point prompts, while latest specialized medical imaging models achieve better performance than general purpose ones. The proposed approach obtains a Dice coefficient of 0.85, suggesting that synthetic prompt generation as a promising strategy for lung nodule segmentation with foundation models.

eess.IV↗

arXiv2026

Global Logic and Local Search: Dual-Stream Multimodal In-Context Learning for Verifiable Industrial Anomaly Detection

Large Multimodal Models (LMMs) show strong few-shot generalization, but industrial anomaly detection remains difficult because defects are small, input resolution is limited, and textual standards are not always grounded in visual evidence. Recent optimization-based methods improve alignment through fine-tuning, but they often require many defective samples, which are unavailable in early deployment. We present Global Logic and Local Search (GLLS), a training-free framework for reference-guided multimodal in-context verification. GLLS uses a Part-Aware Visual-Logical Atlas to organize normal references and structured specifications in the inference context. It combines a Global & Logic Stream, where SAM 3 extracts partially checkable visual facts, with a Fine-Grained & Actions Stream, where MCTS selects local evidence crops under a fixed budget. Experiments on MMAD-QA and additional anomaly detection datasets show consistent gains over matched and general-purpose baselines, while keeping the final diagnostic decision traceable to explicit visual evidence throughout the inspection trace.

cs.CV↗

arXiv2026

Medical Image Understanding Improves Survival Prediction via Visual Instruction Tuning

Accurate prognostication and risk estimation are essential for guiding clinical decision-making and optimizing patient management. While radiologist-assessed features from CT scans provide valuable indicators of disease severity and outcomes, interpreting such images requires expert knowledge, and translating rich visual information into textual summaries inevitably leads to information loss. In this work, we propose a vision-language framework for 3D CT image understanding that leverages large-scale open-sourced CT images paired with radiology reports through visual instruction tuning. This pre-training enables the model to learn clinically meaningful visual-textual representations, which can then be adapted to downstream survival prediction tasks. By incorporating a survival prediction head on top of the pre-trained model, our approach improves survival prediction from CT images and clinical data while generating clinically meaningful language responses to predefined questions. Experimental results demonstrate that our method outperforms baseline methods in survival prediction, particularly, when clinical data alone is less predictive. The code will be released upon acceptance.

cs.CV↗

arXiv2025

ABounD: Adversarial Boundary-Driven Few-Shot Learning for Multi-Class Anomaly Detection

Few-shot multi-class industrial anomaly detection identifies diverse defects across multiple categories using a single unified model and limited normal samples. Although vision-language models offer strong generalization, modeling multiple distinct category manifolds concurrently without actual anomalous data causes feature space collapse and cross-class interference. Consequently, existing methods often fail to balance scalability and precision, leading to either isolated single-class retraining or excessively loose decision margins. To address this limitation, we present a one-for-all learning framework called ABounD that unites semantic concept anchoring with geometric boundary optimization. This method employs two lightweight mechanisms to resolve multi-class ambiguity. First, the Dynamic Concept Fusion module generates class-adaptive semantic anchors via query-aware hierarchical calibration, disentangling overlapping category concepts. Second, using these anchors, the Adversarial Boundary Forging module constructs a tight, class-tailored decision margin by synthesizing adversarial boundary-level fence features to prevent cross-class boundary blurring. Optimized in a single stage, ABounD removes the requirement for isolated per-category retraining in few-shot settings. Experiments on seven industrial benchmarks show that the proposed method achieves state-of-the-art detection and localization performance for multi-class few-shot anomaly detection while maintaining low computational costs during training and inference.

cs.CV↗

arXiv2025

A Transferable Physics-Informed Framework for Battery Degradation Diagnosis, Knee-Onset Detection and Knee Prediction

The techno-economic and safety concerns of battery capacity knee occurrence call for developing online knee detection and prediction methods as an advanced battery management system (BMS) function. To address this, a transferable physics-informed framework that consists of a histogram-based feature engineering method, a hybrid physics-informed model, and a fine-tuning strategy, is proposed for online battery degradation diagnosis and knee-onset detection. The hybrid model is first developed and evaluated using a scenario-aware pipeline in protocol cycling scenarios and then fine-tuned to create local models deployed in a dynamic cycling scenario. A 2D histogram-based 17-feature set is found to be the best choice in both source and target scenarios. The fine-tuning strategy is proven to be effective in improving battery degradation mode estimation and degradation phase detection performance in the target scenario. Again, a strong linear correlation was found between the identified knee-onset and knee points. As a result, advanced BMS functions, such as online degradation diagnosis and prognosis, online knee-onset detection and knee prediction, aging-aware battery classification, and second-life repurposing, can be enabled through a battery performance digital twin in the cloud.

eess.SY↗

arXiv2025

Energy-Guided Decoding for Object Hallucination Mitigation

Mitigating object hallucination in large vision-language models (LVLMs) is critical to their safe deployment. Existing methods either are restricted to specific decoding methods, or demand sophisticated modifications to visual inputs, or rely on knowledge from external models. In this work, we first reveal the phenomenon that VLMs exhibit significant imbalance in the ``Yes'' ratio ( \ie, the fraction of ``Yes'' answers among the total number of questions) across three different visual question answering (VQA) datasets. Furthermore, we propose an energy-based decoding method, which dynamically selects the hidden states from the layer with minimal energy score. It is simple yet effective in reducing the bias for the yes ratio while boosting performance across three benchmarks (POPE, MME, and MMVP). Our method consistently improves accuracy and F1 score on three VQA datasets across three commonly used VLMs over several baseline methods. The average accuracy improvement is 4.82% compared to greedy decoding. Moreover, the average yes-ratio gap reduction is 8.81%, meaning the proposed method is less biased as shown in Figure 1.

cs.CV↗

arXiv2025

A Practitioner's Guide to Automatic Kernel Search for Gaussian Processes in Battery Applications

Gaussian process (GP) models have been used in a wide range of battery applications, in which different kernels were manually selected with considerable expertise. However, to capture complex relationships in the ever-growing amount of real-world data, selecting a suitable kernel for the GP model in battery applications is increasingly challenging. In this work, we first review existing GP kernels used in battery applications and then extend an automatic kernel search method with a new base kernel and model selection criteria. The GP models with composite kernels outperform the baseline kernel in two numerical examples of battery applications, i.e., battery capacity estimation and residual load prediction. Particularly, the results indicate that the Bayesian Information Criterion may be the best model selection criterion as it achieves a good trade-off between kernel performance and computational complexity. This work should, therefore, be of value to practitioners wishing to automate their kernel search process in battery applications.

eess.SY↗

arXiv2025

Enhancing Out-of-Distribution Detection with Extended Logit Normalization

\noindent Out-of-distribution (OOD) detection is essential for the safe deployment of machine learning models. Extensive work has focused on devising various scoring functions for detecting OOD samples, while only a few studies focus on training neural networks using certain model calibration objectives, which often lead to a compromise in predictive accuracy and support only limited choices of scoring functions. In this work, we first identify the feature collapse phenomena in Logit Normalization (LogitNorm), then propose a novel hyperparameter-free formulation that significantly benefits a wide range of post-hoc detection methods. To be specific, we devise a feature distance-awareness loss term in addition to LogitNorm, termed $\textbf{ELogitNorm}$, which enables improved OOD detection and in-distribution (ID) confidence calibration. Extensive experiments across standard benchmarks demonstrate that our approach outperforms state-of-the-art training-time methods in OOD detection while maintaining strong ID classification accuracy. Our code is available on: https://github.com/limchaos/ElogitNorm.

cs.CV↗