arXiv subjects

Ngai-Man Cheung

Publications and source records attributed to Ngai-Man Cheung.

At least 19 recordsLinked to original sources

arXiv2026

Joint Architecture-Token-Bitwidth Multi-Axis Optimization of Vision Transformers for Semiconductor IC Packaging

Vision Transformers (ViTs) have achieved strong visual recognition performance, yet their deployment in resource-constrained industrial environments remains limited. The main challenges are their high computational cost, memory requirements, and energy consumption. While individual efficiency techniques such as neural architecture search (NAS), token compression, and low-precision inference have been extensively studied, most prior work targets only a single optimization axis, limiting overall deployment gains while preserving accuracy. In this paper, we present one of the first holistic frameworks that jointly optimizes three complementary axes: architecture, token, and bit-width. Specifically, the framework identifies compact backbones via Neural Architecture Search (AutoFormer), reduces information processing via token merging (ToMe), and accelerates per-operation execution via fp16 mixed-precision inference. In our study, we analyze accuracy-efficiency trade-offs on ImageNet-1K under aggressive compression. We then apply the selected configurations to a real-world in-house 3D X-ray semiconductor defect classification dataset for IC chip packaging inspection. Results show that the proposed multi-axis framework achieves more than $10\times$ improvement in throughput along with over $10\times$ reductions in parameter count, FLOPs, and energy consumption, while maintaining the required accuracy on the downstream industrial task. To the best of our knowledge, this is among the earliest works to jointly optimize architecture, token, and bit-width dimensions in ViTs and the first such resource-efficient, deployment-focused study for semiconductor manufacturing.

cs.CV↗

arXiv2026

Spatially-Aware Class-Agnostic Object Counting

Generalised object counting aims to estimate the number of instances of an arbitrary object category from a single image, but many recent methods can struggle on structurally complex objects due to limited spatial modelling. We present \textit{UpCount}, a class-agnostic counter designed to better preserve spatial structure. UpCount strengthens the visual representation by extracting multi-layer features from a ViT-B/16 encoder and reassembling them into a refined multi-scale pyramid that is spatially refined using Dense Prediction Transformers and FeatUp, yielding features with improved structural and spatial sensitivity; a proposal--verification counting head then identifies repeated patterns and produces a density map for the final count. On FSC-147, UpCount achieves 12.39 MAE and 100.89 RMSE on the test set, and it transfers effectively to vehicle counting on CARPK (6.27 MAE, 8.79 RMSE). Code: https://github.com/r28112072-rgb/upcount

cs.CV↗

arXiv2026

Mixture of Cognitive Experts in Large Vision-Language Models

Large Vision Language Models (LVLMs) require strong reasoning over both visual and textual input. Recent work suggests that cognitive elements, especially diverse representations and metacognition, correlate with better performance. Many of the needed perceptual functions are already provided by specialized domain-specific computer vision models, which act as the perceptual subsystem for detecting objects, localizing them, inferring states, recovering spatial layout, and reading text. The key challenge is to integrate these multi-encoder experts into a trustworthy, interpretable, and coherent representation that improves verifiability and reduces hallucinations. This is difficult because vision-language questions span different cognitive levels, yet most LVLM pipelines apply the same perception-reasoning routing regardless of the demand of each query. We propose an evidence-driven multimodal reasoning framework that utilizes a Bloom-inspired taxonomy as a hierarchical reasoning protocol. The two-stage cognitive verbalization first produces a Literal Evidence Summary by decomposing expert outputs into short, atomic evidence statements. It then performs Bloom Verbalization to turn these evidence items into a staged reasoning trace, and a lightweight Reasoning Trace Module quantitatively analyzes the trace to make evidence usage and reasoning progression explicit. Through this integration, we observed several improvements in perception and reasoning abilities. Moreover, the trace module provides quantitative evidence that different queries induce different cognitive entry levels and evidence-use trajectories that enable fine-grained analysis.

cs.CV↗

arXiv2026

SAVER: Mitigating Hallucinations in Large Vision-Language Models via Style-Aware Visual Early Revision

Large Vision-Language Models (LVLMs) recently achieve significant breakthroughs in understanding complex visual-textual contexts. However, hallucination issues still limit their real-world applicability. Although previous mitigation methods effectively reduce hallucinations in photographic images, they largely overlook the potential risks posed by stylized images, which play crucial roles in critical scenarios such as game scene understanding, art education, and medical analysis. In this work, we first construct a dataset comprising photographic images and their corresponding stylized versions with carefully annotated caption labels. We then conduct head-to-head comparisons on both discriminative and generative tasks by benchmarking 13 advanced LVLMs on the collected datasets. Our findings reveal that stylized images tend to induce significantly more hallucinations than their photographic counterparts. To address this issue, we propose Style-Aware Visual Early Revision SAVER, a novel mechanism that dynamically adjusts LVLMs' final outputs based on the token-level visual attention patterns, leveraging early-layer feedback to mitigate hallucinations caused by stylized images. Extensive experiments demonstrate that SAVER achieves state-of-the-art performance in hallucination mitigation across various models, datasets, and tasks.

cs.CV↗

arXiv2026

When Token Compression Breaks: Structural Pruning vs. Token Reduction for Robust ViT Segmentation under High Compression

Vision Transformers (ViTs) are strong backbones for semantic segmentation, but their computational cost limits deployment. Recent token compression methods for efficient transformer-based segmentation reduce this cost by decreasing the number of tokens. However, existing evaluations primarily focus on low-to-moderate compression, leaving their behavior under aggressive compression and corrupted inputs unclear. Meanwhile, structural pruning provides an orthogonal route to efficiency by removing redundant components in the ViT architecture, but is rarely compared to token compression under a unified protocol. To bridge this gap, we benchmark representative token compression and structural pruning methods for ViT-based semantic segmentation under matched FLOPs on ADE20K and Cityscapes, together with their common-corruption variants ADE20K-C and Cityscapes-C. Our results reveal a consistent trend on both clean and corrupted inputs: token compression is highly effective at mild reductions but degrades sharply when compression becomes severe, consistent with substantial information loss from overly aggressive token reduction. In contrast, structural pruning exhibits a smoother degradation curve and is more stable at high compression. Motivated by these findings, we study a prune-then-merge pipeline that applies moderate token compression on top of a moderately pruned backbone. At comparable FLOPs, this combined strategy consistently achieves a better accuracy-robustness trade-off at high compression, offering a practical recipe for deployment-oriented ViT segmentation. Code is available at https://github.com/phatnguyencs/vit-seg-compression.

cs.CV↗

arXiv2026

Random Erasing vs. Model Inversion: A Promising Defense or a False Hope?

Model Inversion (MI) attacks pose a significant privacy threat by reconstructing private training data from machine learning models. While existing defenses primarily concentrate on model-centric approaches, the impact of data on MI robustness remains largely unexplored. In this work, we explore Random Erasing (RE), a technique traditionally used for improving model generalization under occlusion, and uncover its surprising effectiveness as a defense against MI attacks. Specifically, our novel feature space analysis shows that models trained with RE-images introduce a significant discrepancy between the features of MI-reconstructed images and those of the private data. At the same time, features of private images remain distinct from other classes and well-separated from different classification regions. These effects collectively degrade MI reconstruction quality and attack accuracy while maintaining reasonable natural accuracy. Furthermore, we explore two critical properties of RE including Partial Erasure and Random Location. Partial Erasure prevents the model from observing entire objects during training. We find this has a significant impact on MI, which aims to reconstruct the entire objects. Random Location of erasure plays a crucial role in achieving a strong privacy-utility trade-off. Our findings highlight RE as a simple yet effective defense mechanism that can be easily integrated with existing privacy-preserving techniques. Extensive experiments across 37 setups demonstrate that our method achieves state-of-the-art (SOTA) performance in the privacy-utility trade-off. The results consistently demonstrate the superiority of our defense over existing methods across different MI attacks, network architectures, and attack configurations. For the first time, we achieve a significant degradation in attack accuracy without a decrease in utility for some configurations.

cs.LG↗

arXiv2026

Characterizing Detectability in 3DGS Poisoning: A Stage-wise Benchmark

3D Gaussian Splatting (3DGS) has rapidly emerged as a leading representation for real-time novel view synthesis, but recent work shows it is vulnerable to diverse poisoning attacks, including illusory object injection, computation cost amplification, and post hoc model watermarking. Despite this expanding threat surface, existing studies focus mainly on attack success, while defense and detection remain underexplored. From a detection perspective, a key challenge and opportunity arise from the multi-stage nature of the 3DGS reconstruction pipeline, which produces heterogeneous intermediate representations. Forensic signals for detecting poisoning are inherently stage dependent: an attack introduced at one stage may produce signals that emerge only at later stages. This motivates a stage-wise view of detectability that goes beyond single-stage evaluation. We introduce Poison-3DGS, a benchmark for stage-wise characterization of poisoning detection in 3DGS. It exposes stage-specific artifacts, including multi-view images, geometry, training dynamics, and Gaussian parameters, across a diverse set of scenes and attacks. Using it, we conduct a systematic study of detectability across pipeline stages. Our analysis reveals several insights. First, detectability varies significantly across stages, and no single stage consistently dominates across attack types. Second, different attacks exhibit distinct stage-specific forensic signals, so detection effectiveness depends critically on where signals are observed. Third, later-stage signals such as training dynamics and Gaussian parameter statistics provide strong cues not observable at earlier stages. Overall, our work provides a principled benchmark and the first systematic characterization of stage-dependent detectability in 3DGS, offering a foundation for future research on robust and reliable 3DGS systems.

cs.CV↗

arXiv2026

Revisiting Model Inversion Evaluation: From Misleading Standards to Reliable Privacy Assessment

Model Inversion attacks aim to reconstruct information from private training data by exploiting access to a target model. Nearly all recent MI studies evaluate attack success using a standard framework that computes attack accuracy through a secondary evaluation model trained on the same private data and task design as the target model. In this paper, we present the first in-depth analysis of this dominant evaluation framework and reveal a fundamental issue: many reconstructions deemed successful under the existing framework are in fact false positives that do not capture the visual identity of the target individual. We first show that these MI false positives satisfy the same formal conditions as Type I adversarial examples. Our controlled experiments, we demonstrate extremely high false-positive transferability, an empirical signature characteristic of adversarial behavior, indicating that many MI false positives likely contain Type I adversarial features. This adversarial transferability significantly inflates reported attack accuracy and leads to an overstatement of privacy leakage in existing MI work. To address this issue, as our second contribution, we introduce a new evaluation framework based on MLLMs, whose general-purpose visual reasoning avoids the shared-task vulnerability and reduces Type-I adversarial transferability of current evaluation framework. We propose systematic design principles for MLLM-based evaluation. Using this framework, we reassess 27 MI attack setups across diverse datasets, target models, and priors, and find consistently high false-positive rates under the conventional approach. Our results call for a reevaluation of progress in MI research and establish MLLM-based evaluation as a more reliable standard for assessing privacy risks in machine learning systems. Code/data/prompt are available at https://hosytuyen.github.io/projects/FMLLM

cs.LG↗

arXiv2026

CERSA: Cumulative Energy-Retaining Subspace Adaptation for Memory-Efficient Fine-Tuning

To mitigate the memory constraints associated with fine-tuning large pre-trained models, existing parameter-efficient fine-tuning (PEFT) methods, such as LoRA, rely on low-rank updates. However, such updates fail to fully capture the rank characteristics of the weight modifications observed in full-parameter fine-tuning, resulting in a performance gap. Furthermore, LoRA and other existing PEFT methods still require substantial memory to store the full set of frozen weights, limiting their efficiency in resource-constrained settings. To addres these limitations, we introduce Cumulative Energy-Retaining Subspace Adaptation (CERSA), a novel fine-tuning paradigm that leverages singular value decomposition (SVD) to retain only the principal components responsible for 90% to 95% of the spectral energy. By fine-tuning low-rank representations derived from this principal subspace, CERSA significantly reduces memory consumption. We conduct extensive evaluations of CERSA across models of varying scales and domains, including image recognition, text-to-image generation, and natural language understanding. Empirical results demonstrate that CERSA consistently outperforms state-of-the-art PEFT methods while achieving substantially lower memory requirements. The code will be publicly released.

cs.LG↗

arXiv2026

How Do Medical MLLMs Fail? A Study on Visual Grounding in Medical Images

Generalist multimodal large language models (MLLMs) have achieved impressive performance across a wide range of vision-language tasks. However, their performance on medical tasks, particularly in zero-shot settings where generalization is critical, remains suboptimal. A key research gap is the limited understanding of why medical MLLMs underperform in medical image interpretation. In this work, we present a pioneering systematic investigation into the visual grounding capabilities of state-of-the-art medical MLLMs. To disentangle visual grounding from semantic grounding, we design VGMED, a novel evaluation dataset developed with expert clinical guidance, explicitly assessing the visual grounding capability of medical MLLMs. We introduce new quantitative metrics and conduct detailed qualitative analyses. Our study across eight state-of-the-art (SOTA) medical MLLMs validates that they often fail to ground their predictions in clinically relevant image regions. We note that this finding is specific to medical image analysis; in contrast, prior work has shown that MLLMs are capable of grounding their predictions in the correct image regions when applied to natural scene images. Motivated by these findings, we propose VGRefine, a simple yet effective inference-time method that refines attention distribution to improve visual grounding in medical settings. Our approach achieves SOTA performance across 6 diverse Med-VQA benchmarks (over 110K VQA samples from 8 imaging modalities) without requiring additional training or external expert models. Overall, our work, for the first time, systematically validates inadequate visual grounding as one of the key contributing factors for medical MLLMs' under-performance. Additional experiments are included in the Supp.

cs.CV↗

arXiv2026

Do Vision-Language Models Leak What They Learn? Adaptive Token-Weighted Model Inversion Attacks

Model inversion (MI) attacks pose significant privacy risks by reconstructing private training data from trained neural networks. While prior studies have primarily examined unimodal deep networks, the vulnerability of vision-language models (VLMs) remains largely unexplored. In this work, we present the first systematic study of MI attacks on VLMs to understand their susceptibility to leaking private visual training data. Our work makes two main contributions. First, tailored to the token-generative nature of VLMs, we introduce a suite of token-based and sequence-based model inversion strategies, providing a comprehensive analysis of VLMs' vulnerability under different attack formulations. Second, based on the observation that tokens vary in their visual grounding, and hence their gradients differ in informativeness for image reconstruction, we propose Sequence-based Model Inversion with Adaptive Token Weighting (SMI-AW) as a novel MI for VLMs. SMI-AW dynamically reweights each token's loss gradient according to its visual grounding, enabling the optimization to focus on visually informative tokens and more effectively guide the reconstruction of private images. Through extensive experiments and human evaluations on a range of state-of-the-art VLMs across multiple datasets, we show that VLMs are susceptible to training data leakage. Human evaluation of the reconstructed images yields an attack accuracy of 61.21%, underscoring the severity of these privacy risks. Notably, we demonstrate that publicly released VLMs are vulnerable to such attacks. Our study highlights the urgent need for privacy safeguards as VLMs become increasingly deployed in sensitive domains such as healthcare and finance. Our code and models are available at our project page: https://ngoc-nguyen-0.github.io/SMI_AW/

cs.LG↗

arXiv2026

A Survey on Generative Modeling with Limited Data, Few Shots, and Zero Shot

Generative modeling in machine learning aims to synthesize new data samples that are statistically similar to those observed during training. While conventional generative models such as GANs and diffusion models typically assume access to large and diverse datasets, many real-world applications (e.g. in medicine, satellite imaging, and artistic domains) operate under limited data availability and strict constraints. In this survey, we examine Generative Modeling under Data Constraint (GM-DC), which includes limited-data, few-shot, and zero-shot settings. We present a unified perspective on the key challenges in GM-DC, including overfitting, frequency bias, and incompatible knowledge transfer, and discuss how these issues impact model performance. To systematically analyze this growing field, we introduce two novel taxonomies: one categorizing GM-DC tasks (e.g. unconditional vs. conditional generation, cross-domain adaptation, and subject-driven modeling), and another organizing methodological approaches (e.g. transfer learning, data augmentation, meta-learning, and frequency-aware modeling). Our study reviews over 230 papers, offering a comprehensive view across generative model types and constraint scenarios. We further analyze task-approach-method interactions using a Sankey diagram and highlight promising directions for future work, including adaptation of foundation models, holistic evaluation frameworks, and data-centric strategies for sample selection. This survey provides a timely and practical roadmap for researchers and practitioners aiming to advance generative modeling under limited data. Project website: https://sutd-visual-computing-group.github.io/gmdc-survey/.

cs.CV↗

arXiv2026

Towards Sustainable Universal Deepfake Detection with Frequency-Domain Masking

Universal deepfake detection aims to identify AI-generated images across a broad range of generative models, including unseen ones. This requires robust generalization to new and unseen deepfakes, which emerge frequently, while minimizing computational overhead to enable large-scale deepfake screening, a critical objective in the era of Green AI. In this work, we explore frequency-domain masking as a training strategy for deepfake detectors. Unlike traditional methods that rely heavily on spatial features or large-scale pretrained models, our approach introduces random masking and geometric transformations, with a focus on frequency masking due to its superior generalization properties. We demonstrate that frequency masking not only enhances detection accuracy across diverse generators but also maintains performance under significant model pruning, offering a scalable and resource-conscious solution. Our method achieves state-of-the-art generalization on GAN- and diffusion-generated image datasets and exhibits consistent robustness under structured pruning. These results highlight the potential of frequency-based masking as a practical step toward sustainable and generalizable deepfake detection. Code and models are available at https://github.com/chandlerbing65nm/FakeImageDetection.

cs.CV↗

arXiv2026

On the Adversarial Robustness of 3D Large Vision-Language Models

3D Vision-Language Models (VLMs), such as PointLLM and GPT4Point, have shown strong reasoning and generalization abilities in 3D understanding tasks. However, their adversarial robustness remains largely unexplored. Prior work in 2D VLMs has shown that the integration of visual inputs significantly increases vulnerability to adversarial attacks, making these models easier to manipulate into generating toxic or misleading outputs. In this paper, we investigate whether incorporating 3D vision similarly compromises the robustness of 3D VLMs. To this end, we present the first systematic study of adversarial robustness in point-based 3D VLMs. We propose two complementary attack strategies: \textit{Vision Attack}, which perturbs the visual token features produced by the 3D encoder and projector to assess the robustness of vision-language alignment; and \textit{Caption Attack}, which directly manipulates output token sequences to evaluate end-to-end system robustness. Each attack includes both untargeted and targeted variants to measure general vulnerability and susceptibility to controlled manipulation. Our experiments reveal that 3D VLMs exhibit significant adversarial vulnerabilities under untargeted attacks, while demonstrating greater resilience against targeted attacks aimed at forcing specific harmful outputs, compared to their 2D counterparts. These findings highlight the importance of improving the adversarial robustness of 3D VLMs, especially as they are deployed in safety-critical applications.

cs.CV↗

arXiv2025

Synth-Align: Improving Trustworthiness in Vision-Language Model with Synthetic Preference Data Alignment

Large Vision-Language Models (LVLMs) have shown promising capabilities in understanding and generating information by integrating both visual and textual data. However, current models are still prone to hallucinations, which degrade the performance and greatly harm the user experience in real-world applications. Post-training alignment, particularly preference-tuning, is intended to align model outputs and behaviors (safety, instruction-following, style), ensuring robustness and adaptability to a wide range of tasks. The use of synthetic data for alignment, particularly in multimodal settings, remains under explored. Existing approaches typically use a strong model or a ground-truth model (CLIP) to determine positive and negative image-text data points. This paper proposes SynthAlign, a pipeline to generate and collect synthetic human-preference image-text data with optimal control built specifically for post-training alignment with DPO. At the core of the framework is the utilization of reward models as a proxy of human preference. A series of evaluation and benchmarking is provided to validate the effectiveness of the proposed framework and the resulting dataset. Notably, our framework enhanced LLaVA-1.5-7B achieved substantial POPE improvements: 87.6\% accuracy and 97.8\% precision, MMHal-Bench score increased from 2.36 to 3.49, and hallucination rate decreased from 51.0\% to 25.0\% (a 50.98\% relative reduction).

cs.CV↗

arXiv2025

Token Compression Meets Compact Vision Transformers: A Survey and Comparative Evaluation for Edge AI

Token compression techniques have recently emerged as powerful tools for accelerating Vision Transformer (ViT) inference in computer vision. Due to the quadratic computational complexity with respect to the token sequence length, these methods aim to remove less informative tokens before the attention layers to improve inference throughput. While numerous studies have explored various accuracy-efficiency trade-offs on large-scale ViTs, two critical gaps remain. First, there is a lack of unified survey that systematically categorizes and compares token compression approaches based on their core strategies (e.g., pruning, merging, or hybrid) and deployment settings (e.g., fine-tuning vs. plug-in). Second, most benchmarks are limited to standard ViT models (e.g., ViT-B, ViT-L), leaving open the question of whether such methods remain effective when applied to structurally compressed transformers, which are increasingly deployed on resource-constrained edge devices. To address these gaps, we present the first systematic taxonomy and comparative study of token compression methods, and we evaluate representative techniques on both standard and compact ViT architectures. Our experiments reveal that while token compression methods are effective for general-purpose ViTs, they often underperform when directly applied to compact designs. These findings not only provide practical insights but also pave the way for future research on adapting token optimization techniques to compact transformer-based networks for edge AI and AI agent applications.

cs.CV↗

arXiv2025

AIR: Zero-shot Generative Model Adaptation with Iterative Refinement

Zero-shot generative model adaptation (ZSGM) aims to adapt a pre-trained generator to a target domain using only text guidance and without any samples from the target domain. Central to recent ZSGM approaches are directional loss which use the text guidance in the form of aligning the image offset with text offset in the embedding space of a vision-language model like CLIP. This is similar to the analogical reasoning in NLP where the offset between one pair of words is used to identify a missing element in another pair by aligning the offset between these two pairs. However, a major limitation of existing ZSGM methods is that the learning objective assumes the complete alignment between image offset and text offset in the CLIP embedding space, resulting in quality degrade in generated images. Our work makes two main contributions. Inspired by the offset misalignment studies in NLP, as our first contribution, we perform an empirical study to analyze the misalignment between text offset and image offset in CLIP embedding space for various large publicly available datasets. Our important finding is that offset misalignment in CLIP embedding space is correlated with concept distance, i.e., close concepts have a less offset misalignment. To address the limitations of the current approaches, as our second contribution, we propose Adaptation with Iterative Refinement (AIR) which is the first ZSGM approach to focus on improving target domain image quality based on our new insight on offset misalignment.Qualitative, quantitative, and user study in 26 experiment setups consistently demonstrate the proposed AIR approach achieves SOTA performance. Additional experiments are in Supp.

cs.CV↗

arXiv2025

Text to Image Generation and Editing: A Survey

Text-to-image generation (T2I) refers to the text-guided generation of high-quality images. In the past few years, T2I has attracted widespread attention and numerous works have emerged. In this survey, we comprehensively review 141 works conducted from 2021 to 2024. First, we introduce four foundation model architectures of T2I (autoregression, non-autoregression, GAN and diffusion) and the commonly used key technologies (autoencoder, attention and classifier-free guidance). Secondly, we systematically compare the methods of these studies in two directions, T2I generation and T2I editing, including the encoders and the key technologies they use. In addition, we also compare the performance of these researches side by side in terms of datasets, evaluation metrics, training resources, and inference speed. In addition to the four foundation models, we survey other works on T2I, such as energy-based models and recent Mamba and multimodality. We also investigate the potential social impact of T2I and provide some solutions. Finally, we propose unique insights of improving the performance of T2I models and possible future development directions. In summary, this survey is the first systematic and comprehensive overview of T2I, aiming to provide a valuable guide for future researchers and stimulate continued progress in this field.

cs.CV↗