arXiv subjects

Yimu Pan

Publications and source records attributed to Yimu Pan.

9 recordsLinked to original sources

arXiv2026

Shieldstral

We introduce Shieldstral, a 3B-parameter policy-adaptive multimodal safety classifier that matches or outperforms models nearly 7$\times$ its size on text safety benchmarks and sets a new state of the art on multimodal safety classification. Shieldstral formulates content moderation as a binary question-answering task. This simple formulation unifies diverse moderation tasks into a single yes/no problem, enabling heterogeneous safety datasets with divergent taxonomies to be consolidated under one training framework. We present the data construction recipe, covering curation and generation of approximately 54.1M samples and a fine-grained evaluation set to evaluate policy adaptability. Together, these enable a small adaptive model to match or outperform much larger models.

cs.CL↗

arXiv2026

Phoenix-VL 1.5 Medium Technical Report

We introduce Phoenix-VL 1.5 Medium, a 123B-parameter natively multimodal and multilingual foundation model, adapted to regional languages and the Singapore context. Developed as a sovereign AI asset, it demonstrates that deep domain adaptation can be achieved with minimal degradation to broad-spectrum intelligence and alignment. Continued pretraining was performed on Mistral Medium 3.1 using a localized 1-trillion tokens multimodal corpus, followed by a 250-billion tokens long-context extension phase. Subsequent post-training incorporated a novel human-annotated Singapore multimodal dataset and curated textual corpus on Singapore culture, knowledge, and legislation, totaling 22-billion tokens. An additional 5 billion tokens of model alignment was performed through Online Direct Preference Optimization. Phoenix-VL 1.5 Medium achieves state-of-the-art performance for its size on Singapore multimodal, legal, and government policy benchmarks while remaining globally competitive on general multimodal intelligence, multilingual, and STEM benchmarks. We also introduce a novel evaluation suite encompassing localized knowledge benchmarks and an institutionally aligned model behavior and safety framework. We report the data curation principles, training methodology, and highlight benchmark and inference performance.

cs.CL↗

arXiv2026

studentSplat: Your Student Model Learns Single-view 3D Gaussian Splatting

Recent advance in feed-forward 3D Gaussian splatting has enable remarkable multi-view 3D scene reconstruction or single-view 3D object reconstruction but single-view 3D scene reconstruction remain under-explored due to inherited ambiguity in single-view. We present \textbf{studentSplat}, a single-view 3D Gaussian splatting method for scene reconstruction. To overcome the scale ambiguity and extrapolation problems inherent in novel-view supervision from a single input, we introduce two techniques: 1) a teacher-student architecture where a multi-view teacher model provides geometric supervision to the single-view student during training, addressing scale ambiguity and encourage geometric validity; and 2) an extrapolation network that completes missing scene context, enabling high-quality extrapolation. Extensive experiments show studentSplat achieves state-of-the-art single-view novel-view reconstruction quality and comparable performance to multi-view methods at the scene level. Furthermore, studentSplat demonstrates competitive performance as a self-supervised single-view depth estimation method, highlighting its potential for general single-view 3D understanding tasks.

cs.CV↗

arXiv2025

RulerNet: Learning Perspective-Invariant Ruler Representations for Robust Image Scale Estimation

Accurately converting pixel measurements into absolute real-world dimensions remains a fundamental challenge in computer vision, limiting progress in applications such as biomedicine, forensics, nutritional analysis, and e-commerce. We introduce RulerNet, a deep learning framework that robustly infers scale in the wild by reformulating ruler reading as a unified keypoint detection problem and representing rulers with geometric progression parameters that compactly approximate the non-uniform spacing induced by perspective transformations. Unlike traditional methods that rely on handcrafted thresholds or rigid, ruler-specific pipelines, RulerNet directly localizes centimeter markings using a mark-visibility-based annotation and training strategy that remains valid under mark-preserving transformations, enabling strong generalization across diverse ruler types and imaging conditions while mitigating data scarcity. Additionally, we introduce a scalable synthetic data generation pipeline that combines graphics-based ruler creation with ControlNet-enhanced realism, significantly expanding training diversity and improving model performance. Extensive experiments on both our datasets and public benchmarks demonstrate that RulerNet achieves accurate, consistent, and efficient scale estimation under challenging real-world conditions. Integration into a medical analysis pipeline further demonstrates its practical utility for scale-aware measurement. These results suggest that RulerNet can serve as a generalizable measurement component and be readily integrated with other vision modules, such as segmentation, monocular depth estimation, and diagnostic workflows, for automated analysis in medical and other domains. An online demo is provided. The project page is at https://github.com/ymp5078/RulerNet.

cs.CV↗

arXiv2025

VLCD: Vision-Language Contrastive Distillation for Accurate and Efficient Automatic Placenta Analysis

Pathological examination of the placenta is an effective method for detecting and mitigating health risks associated with childbirth. Recent advancements in AI have enabled the use of photographs of the placenta and pathology reports for detecting and classifying signs of childbirth-related pathologies. However, existing automated methods are computationally extensive, which limits their deployability. We propose two modifications to vision-language contrastive learning (VLC) frameworks to enhance their accuracy and efficiency: (1) text-anchored vision-language contrastive knowledge distillation (VLCD)-a new knowledge distillation strategy for medical VLC pretraining, and (2) unsupervised predistillation using a large natural images dataset for improved initialization. Our approach distills efficient neural networks that match or surpass the teacher model in performance while achieving model compression and acceleration. Our results showcase the value of unsupervised predistillation in improving the performance and robustness of our approach, specifically for lower-quality images. VLCD serves as an effective way to improve the efficiency and deployability of medical VLC approaches, making AI-based healthcare solutions more accessible, especially in resource-constrained environments.

cs.CV↗

arXiv2024

S2S2: Semantic Stacking for Robust Semantic Segmentation in Medical Imaging

Robustness and generalizability in medical image segmentation are often hindered by scarcity and limited diversity of training data, which stands in contrast to the variability encountered during inference. While conventional strategies -- such as domain-specific augmentation, specialized architectures, and tailored training procedures -- can alleviate these issues, they depend on the availability and reliability of domain knowledge. When such knowledge is unavailable, misleading, or improperly applied, performance may deteriorate. In response, we introduce a novel, domain-agnostic, add-on, and data-driven strategy inspired by image stacking in image denoising. Termed ``semantic stacking,'' our method estimates a denoised semantic representation that complements the conventional segmentation loss during training. This method does not depend on domain-specific assumptions, making it broadly applicable across diverse image modalities, model architectures, and augmentation techniques. Through extensive experiments, we validate the superiority of our approach in improving segmentation performance under diverse conditions. Code is available at https://github.com/ymp5078/Semantic-Stacking.

cs.CV↗

arXiv2023

AI-SAM: Automatic and Interactive Segment Anything Model

Semantic segmentation is a core task in computer vision. Existing methods are generally divided into two categories: automatic and interactive. Interactive approaches, exemplified by the Segment Anything Model (SAM), have shown promise as pre-trained models. However, current adaptation strategies for these models tend to lean towards either automatic or interactive approaches. Interactive methods depend on prompts user input to operate, while automatic ones bypass the interactive promptability entirely. Addressing these limitations, we introduce a novel paradigm and its first model: the Automatic and Interactive Segment Anything Model (AI-SAM). In this paradigm, we conduct a comprehensive analysis of prompt quality and introduce the pioneering Automatic and Interactive Prompter (AI-Prompter) that automatically generates initial point prompts while accepting additional user inputs. Our experimental results demonstrate AI-SAM's effectiveness in the automatic setting, achieving state-of-the-art performance. Significantly, it offers the flexibility to incorporate additional user prompts, thereby further enhancing its performance. The project page is available at https://github.com/ymp5078/AI-SAM.

cs.CV↗

arXiv2023

Learning Emotion Representations from Verbal and Nonverbal Communication

Emotion understanding is an essential but highly challenging component of artificial general intelligence. The absence of extensively annotated datasets has significantly impeded advancements in this field. We present EmotionCLIP, the first pre-training paradigm to extract visual emotion representations from verbal and nonverbal communication using only uncurated data. Compared to numerical labels or descriptions used in previous methods, communication naturally contains emotion information. Furthermore, acquiring emotion representations from communication is more congruent with the human learning process. We guide EmotionCLIP to attend to nonverbal emotion cues through subject-aware context encoding and verbal emotion cues using sentiment-guided contrastive learning. Extensive experiments validate the effectiveness and transferability of EmotionCLIP. Using merely linear-probe evaluation protocol, EmotionCLIP outperforms the state-of-the-art supervised visual emotion recognition methods and rivals many multimodal approaches across various benchmarks. We anticipate that the advent of EmotionCLIP will address the prevailing issue of data scarcity in emotion understanding, thereby fostering progress in related domains. The code and pre-trained models are available at https://github.com/Xeaver/EmotionCLIP.

cs.CV↗

arXiv2023

Learning to Adapt to Online Streams with Distribution Shifts

Test-time adaptation (TTA) is a technique used to reduce distribution gaps between the training and testing sets by leveraging unlabeled test data during inference. In this work, we expand TTA to a more practical scenario, where the test data comes in the form of online streams that experience distribution shifts over time. Existing approaches face two challenges: reliance on a large test data batch from the same domain and the absence of explicitly modeling the continual distribution evolution process. To address both challenges, we propose a meta-learning approach that teaches the network to adapt to distribution-shifting online streams during meta-training. As a result, the trained model can perform continual adaptation to distribution shifts in testing, regardless of the batch size restriction, as it has learned during training. We conducted extensive experiments on benchmarking datasets for TTA, incorporating a broad range of online distribution-shifting settings. Our results showed consistent improvements over state-of-the-art methods, indicating the effectiveness of our approach. In addition, we achieved superior performance in the video segmentation task, highlighting the potential of our method for real-world applications.

cs.LG↗