论文检索

Ruiqi Du, Xu Tang, Xiangrong Zhang, Jingjing Ma

Since real-world multi-label data often exhibit significant label imbalance, long-tailed multi-label image classification has emerged as a prominent research area in computer vision. Traditionally, it is considered that deep neural networks' classifiers are vulnerable to long-tailed distributions, whereas the feature extraction backbone remains relatively robust. However, our analysis from the feature learning perspective reveals that the backbone struggles to maintain high sensitivity to sample-scarce categories but retains the ability to localize specific areas effectively. Based on this observation, we propose a new model for long-tailed multi-label image classification named category-specific selective feature enhancement (CSSFE). First, it utilizes the retained localization capability of the backbone to capture label-dependent class activation maps. Then, a progressive attention enhancement mechanism, updating from head to medium to tail categories, is introduced to address the low-confidence issue in medium and tail categories. Finally, visual features are extracted according to the optimized class activation maps and combined with semantic information to perform the classification task. Extensive experiments on two benchmark datasets highlight our findings' generalizability and the proposed CSSFE's superior performance.

Yunhao Li, Yifan Jiao, Dan Meng, Heng Fan, Libo Zhang

Open-Vocabulary Multi-Object Tracking (OV-MOT) aims to enable approaches to track objects without being limited to a predefined set of categories. Current OV-MOT methods typically rely primarily on instance-level detection and association, often overlooking trajectory information that is unique and essential for object tracking tasks. Utilizing trajectory information can enhance association stability and classification accuracy, especially in cases of occlusion and category ambiguity, thereby improving adaptability to novel classes. Thus motivated, in this paper we propose TRACT, an open-vocabulary tracker that leverages trajectory information to improve both object association and classification in OV-MOT. Specifically, we introduce a Trajectory Consistency Reinforcement (TCR) strategy, that benefits tracking performance by improving target identity and category consistency. In addition, we present TraCLIP, a plug-and-play trajectory classification module. It integrates Trajectory Feature Aggregation (TFA) and Trajectory Semantic Enrichment (TSE) strategies to fully leverage trajectory information from visual and language perspectives for enhancing the classification results. Extensive experiments on OV-TAO show that our TRACT significantly improves tracking performance, highlighting trajectory information as a valuable asset for OV-MOT. We will release TRACT at https://github.com/Nathan-Li123/TRACT.

Lu Chen, Yizhou Wang, Shixiang Tang, Qianhong Ma, Tong He, Wanli Ouyang, Xiaowei Zhou, Hujun Bao, Sida Peng

Learning an agent model that behaves like humans--capable of jointly perceiving the environment, predicting the future, and taking actions from a first-person perspective--is a fundamental challenge in computer vision. Existing methods typically train separate models for these abilities, which fail to capture their intrinsic relationships and prevent them from learning from each other. Inspired by how humans learn through the perception-action loop, we propose EgoAgent, a unified agent model that simultaneously learns to represent, predict, and act within a single transformer. EgoAgent explicitly models the causal and temporal dependencies among these abilities by formulating the task as an interleaved sequence of states and actions. It further introduces a joint embedding-action-prediction architecture with temporally asymmetric predictor and observer branches, enabling synergistic optimization across all three capabilities. Comprehensive evaluations of EgoAgent on representative tasks such as image classification, egocentric future state prediction, and 3D human motion prediction demonstrate the superiority of our method. The code and trained models will be publicly available at https://github.com/zju3dv/EgoAgent.

Fatemeh Ghezloo, Mehmet Saygin Seyfioglu, Rustin Soraki, Wisdom O. Ikezogwo, Beibin Li, Tejoram Vivekanandan, Joann G. Elmore, Ranjay Krishna, Linda Shapiro

Diagnosing diseases through histopathology whole slide images (WSIs) is fundamental in modern pathology but is challenged by the gigapixel scale and complexity of WSIs. Trained histopathologists overcome this challenge by navigating the WSI, looking for relevant patches, taking notes, and compiling them to produce a final holistic diagnostic. Traditional AI approaches, such as multiple instance learning and transformer-based models, fail short of such a holistic, iterative, multi-scale diagnostic procedure, limiting their adoption in the real-world. We introduce PathFinder, a multi-modal, multi-agent framework that emulates the decision-making process of expert pathologists. PathFinder integrates four AI agents, the Triage Agent, Navigation Agent, Description Agent, and Diagnosis Agent, that collaboratively navigate WSIs, gather evidence, and provide comprehensive diagnoses with natural language explanations. The Triage Agent classifies the WSI as benign or risky; if risky, the Navigation and Description Agents iteratively focus on significant regions, generating importance maps and descriptive insights of sampled patches. Finally, the Diagnosis Agent synthesizes the findings to determine the patient's diagnostic classification. Our Experiments show that PathFinder outperforms state-of-the-art methods in skin melanoma diagnosis by 8% while offering inherent explainability through natural language descriptions of diagnostically relevant patches. Qualitative analysis by pathologists shows that the Description Agent's outputs are of high quality and comparable to GPT-4o. PathFinder is also the first AI-based system to surpass the average performance of pathologists in this challenging melanoma classification task by 9%, setting a new record for efficient, accurate, and interpretable AI-assisted diagnostics in pathology. Our Data, code and models will be made available.

Benjin Zhu, Xiaogang Wang, Hongsheng Li

Scene synthesis plays a crucial role in autonomous driving by addressing data scarcity and close-loop validation. Current approaches struggle to maintain temporal consistency in synthesized videos while preserving fine-grained details. We introduce ConsistentCity, a two-stage framework with a novel Semantic Flow-guided Diffusion Transformers (SF-DiT) that convert sequential BEV semantic maps into temporally consistent driving videos. Operating in a pretrained occupancy VQ-VAE latent space, our SF-DiT generates temporally consistent 3D occupancy, which provides guidance for controlled image and video diffusion for scene synthesis. To address the temporal consistency, SF-DiT enhances standard DiT blocks with temporal semantic modeling through two designs: (1) A Semantic Flow Estimation module capturing scene motions (flow, uncertainty, and classification) from sequential BEV semantic maps, and (2) A Semantic Flow-Modulated Cross-Attention module that dynamically adapts attention based on semantic flow patterns. This integration of semantic flow modeling in DiT enables consistent scene evolution understanding. Evaluations of image and video synthesis on nuScenes dataset demonstrate state-of-the-art performance with FID 8.3 and FVD 73.6, and superior temporal occupancy generation results on nuCraft and OpenOccupancy benchmarks.

Mian Zou, Nan Zhong, Baosheng Yu, Yibing Zhan, Kede Ma

AI-generated face detectors trained via supervised learning typically rely on synthesized images from specific generators, limiting their generalization to emerging generative techniques. To overcome this limitation, we introduce a self-supervised method based on bi-level optimization. In the inner loop, we pretrain a vision encoder only on photographic face images using a set of linearly weighted pretext tasks: classification of categorical exchangeable image file format (EXIF) tags, ranking of ordinal EXIF tags, and detection of artificial face manipulations. The outer loop then optimizes the relative weights of these pretext tasks to enhance the coarse-grained detection of manipulated faces, serving as a proxy task for identifying AI-generated faces. In doing so, it aligns self-supervised learning more closely with the ultimate goal of AI-generated face detection. Once pretrained, the encoder remains fixed, and AI-generated faces are detected either as anomalies under a Gaussian mixture model fitted to photographic face features or by a lightweight two-layer perceptron serving as a binary classifier. Extensive experiments demonstrate that our detectors significantly outperform existing approaches in both one-class and binary classification settings, exhibiting strong generalization to unseen generators.

Hai Wu, Hongwei Lin, Xusheng Guo, Xin Li, Mingming Wang, Cheng Wang, Chenglu Wen

The performance of unsupervised 3D object classification and bounding box regression relies heavily on the quality of initial pseudo-labels. Traditionally, the labels of classification and regression are represented by a single set of candidate boxes generated by motion or geometry heuristics. However, due to the similarity of many objects to the background in shape or lack of motion, the labels often fail to achieve high accuracy in two tasks simultaneously. Using these labels to directly train the network results in decreased detection performance. To address this challenge, we introduce Motal that performs unsupervised 3D object detection by Modality and task-specific knowledge transfer. Motal decouples the pseudo-labels into two sets of candidates, from which Motal discovers classification knowledge by motion and image appearance prior, and discovers box regression knowledge by geometry prior, respectively. Motal finally transfers all knowledge to a single student network by a TMT (Task-specific Masked Training) scheme, attaining high performance in both classification and regression. Motal can greatly enhance various unsupervised methods by about 2x mAP. For example, on the WOD test set, Motal improves the state-of-the-art CPD by 21.56% mAP L1 (from 20.54% to 42.10%) and 19.90% mAP L2 (from 18.18% to 38.08%). These achievements highlight the significance of our method.

Aryan Yazdan Parast, Basim Azam, Naveed Akhtar

Deep neural networks trained with Empirical Risk Minimization (ERM) perform well when both training and test data come from the same domain, but they often fail to generalize to out-of-distribution samples. In image classification, these models may rely on spurious correlations that often exist between labels and irrelevant features of images, making predictions unreliable when those features do not exist. We propose a Diffusion Driven Balancing (DDB) technique to generate training samples with text-to-image diffusion models for addressing the spurious correlation problem. First, we compute the best describing token for the visual features pertaining to the causal components of samples by a textual inversion mechanism. Then, leveraging a language segmentation method and a diffusion model, we generate new samples by combining the causal component with the elements from other classes. We also meticulously prune the generated samples based on the prediction probabilities and attribution scores of the ERM model to ensure their correct composition for our objective. Finally, we retrain the ERM model on our augmented dataset. This process reduces the model's reliance on spurious correlations by learning from carefully crafted samples in which this correlation does not exist. Our experiments show that across different benchmarks, our technique achieves better worst-group accuracy than the existing state-of-the-art methods. Our code is available at https://github.com/ArianYp/DDB.

Michihiro Kuroki, Toshihiko Yamasaki

Although saliency maps can highlight important regions to explain the reasoning behind image classification in artificial intelligence (AI), the meaning of these regions is left to the user's interpretation. In contrast, concept-based explanations decompose AI predictions into human-understandable concepts, clarifying their contributions. However, few methods can simultaneously reveal what concepts an image classifier learns, which regions are associated with them, and how they contribute to predictions.We propose a novel concept-based explanation method, Concept-based Explanation via Fusion of Activation Maps (CE-FAM). It employs a branched network that shares activation maps with an image classifier and learns to mimic the embeddings of a Vision and Language Model (VLM). The branch network predicts concepts in an image, and their corresponding regions are represented by a weighted sum of activation maps, with weights given by the gradients of the concept prediction scores. Their contributions are quantified based on their impact on the image classification score. Our method provides a general framework for identifying the concept regions and their contributions while leveraging VLM knowledge to handle arbitrary concepts without requiring an annotated dataset. Furthermore, we introduce a novel evaluation metric to assess the accuracy of the concept regions. Our qualitative and quantitative evaluations demonstrate our method outperforms existing approaches and excels in zero-shot inference for unseen concepts.

Jiaqi Wu, Simin Chen, Jing Tang, Yuzhe Yang, Yiming Chen, Lixu Wang, Song Lin, Zehua Wang, Wei Chen, Zijian Tian

General-purpose Vision-Language Models (VLMs) have driven major advancements in multimodal AI. Fine-tuning these models with task-specific data enhances adaptability to various downstream tasks but suffers from privacy risks. While potential solutions like federated learning can address user data privacy concerns, model protection is also essential. Other methods that rely on black-box VLM APIs usually require the access of prediction logits, leaving them open to inversion attacks. Moreover, addressing the challenges of tuning complexity and data transmission efficiency in federated VLM scenarios is also crucial. To address these challenges, we propose FDPT--a federated discrete prompt tuning method utilizing black-box VLMs. During client optimization stage, FDPT employs an agent-driven framework leveraging large language models (LLMs) with enhanced reasoning capacities to systematically optimize discrete prompt representations, and also utilizes feedback mechanisms and chain of thought to enhance prediction accuracy. Importantly, it performs optimization by relying not on the predicted logic vectors output by LLMs but on textual results, avoiding reverse attack risks. During global aggregation stage, We mimic human electoral activities by employing evolutionary computation methods underpinned by semantic similarity computation to implement enhanced zero-order optimization for acquiring representative global tokens, thereby achieving knowledge aggregation. FDPT significantly outperforms nine state-of-the-art methods in image classification and visual question-answering, reducing communication overhead while generating highly transferable optimized prompts. Additionally, it exhibits improved robustness to data heterogeneity.

Srikumar Sastry, Aayush Dhakal, Eric Xing, Subash Khanal, Nathan Jacobs

Learning the hierarchical structure of data in vision-language models is a significant challenge. Previous works have attempted to address this challenge by employing entailment learning. However, these approaches fail to model the transitive nature of entailment explicitly, which establishes the relationship between order and semantics within a representation space. In this work, we introduce Radial Cross-Modal Embeddings (RCME), a framework that enables the explicit modeling of transitivity-enforced entailment. Our proposed framework optimizes for the partial order of concepts within vision-language models. By leveraging our framework, we develop a hierarchical vision-language foundation model capable of representing the hierarchy in the Tree of Life. Our experiments on hierarchical species classification and hierarchical retrieval tasks demonstrate the enhanced performance of our models compared to the existing state-of-the-art models. Our code and models are open-sourced at https://vishu26.github.io/RCME.

Alexey Kravets, Da Chen, Vinay P. Namboodiri

CLIP is a foundational model with transferable classification performance in the few-shot setting. Several methods have shown improved performance of CLIP using few-shot examples. However, so far all these techniques have been benchmarked using standard few-shot datasets. We argue that this mode of evaluation does not provide a true indication of the inductive generalization ability using few-shot examples. As most datasets have been seen by the CLIP model, the resultant setting can be termed as partially transductive. To solve this, we propose a pipeline that uses an unlearning technique to obtain true inductive baselines. In this new inductive setting, methods show a significant drop in performance (-55% on average among 13 baselines with multiple datasets). We validate the unlearning technique using oracle baselines. An improved few-shot classification technique is proposed that consistently obtains state-of-the-art performance over 13 other recent baseline methods on a comprehensive analysis with 5880 experiments - varying the datasets, differing number of few-shot examples, unlearning setting, and with different seeds. Thus, we identify the issue with the evaluation of CLIP-based few-shot classification, provide a solution using unlearning, propose new benchmarks, and provide an improved method.

Mohammed Rakib, Arunkumar Bagavathi

Multimodal learning aims to leverage information from diverse data modalities to achieve more comprehensive performance. However, conventional multimodal models often suffer from modality imbalance, where one or a few modalities dominate model optimization, leading to suboptimal feature representation and underutilization of weak modalities. To address this challenge, we introduce Gradient-Guided Distillation (G^ 2 D), a knowledge distillation framework that optimizes the multimodal model with a custom-built loss function that fuses both unimodal and multimodal objectives. G^ 2 D further incorporates a dynamic sequential modality prioritization (SMP) technique in the learning process to ensure each modality leads the learning process, avoiding the pitfall of stronger modalities overshadowing weaker ones. We validate G^ 2 D on multiple real-world datasets and show that G^ 2 D amplifies the significance of weak modalities while training and outperforms state-of-the-art methods in classification and regression tasks. Our code is available \href https://github.com/rAIson-Lab/G2D here .

Ihab Asaad, Maha Shadaydeh, Joachim Denzler

Machine learning classification models trained with empirical risk minimization (ERM) often inadvertently rely on spurious correlations. When absent in the test data, these unintended associations between non-target attributes and target labels lead to poor generalization. This paper addresses this problem from a model optimization perspective and proposes a novel method, Gradient Extrapolation for Debiased Representation Learning (GERNE), designed to learn debiased representations in both known and unknown attribute training cases. GERNE uses two distinct batches with different amounts of spurious correlations and defines the target gradient as a linear extrapolation of the gradients computed from each batch's loss. Our analysis shows that when the extrapolated gradient points toward the batch gradient with fewer spurious correlations, it effectively guides training toward learning a debiased model. GERNE serves as a general framework for debiasing, encompassing ERM and Resampling methods as special cases. We derive the theoretical upper and lower bounds of the extrapolation factor employed by GERNE. By tuning this factor, GERNE can adapt to maximize either Group-Balanced Accuracy (GBA) or Worst-Group Accuracy (WGA). We validate GERNE on five vision and one NLP benchmarks, demonstrating competitive and often superior performance compared to state-of-the-art baselines. The project page is available at: https://gerne-debias.github.io/.

Wenlong Luo, Shizhou Zhang, De Cheng, Yinghui Xing, Guoqiang Liang, Peng Wang, Yanning Zhang

Incremental object detection (IOD) is crucial for enabling AI systems to continuously learn new object classes over time while retaining knowledge of previously learned categories, allowing model to adapt to dynamic environments without forgetting prior information.Existing IOD methods primarily employ knowledge distillation to mitigate catastrophic forgetting, yet these approaches overlook class overlap issues, often resulting in suboptimal performance. In this paper, we propose a novel framework for IOD that leverages a decoupled gradient alignment technique on top of the specially proposed pseudo-labeling strategy. Our method employs a Gaussian Mixture Model to accurately estimate pseudo-labels of previously learned objects in current training images, effectively functioning as a knowledge-replay mechanism. This strategy reinforces prior knowledge retention and prevents the misclassification of unannotated foreground objects from earlier classes as background. Furthermore, we introduce an adaptive gradient decomposition and alignment method to maintain model stability while facilitating positive knowledge transfer. By aligning gradients from both old and new classes, our approach preserves previously learned knowledge while enhancing plasticity for new tasks. Extensive experiments on two IOD benchmarks demonstrate the effectiveness of the proposed method, achieving superior performances to state-of-the-art methods.

Yiting Yang, Hao Luo, Yuan Sun, Qingsen Yan, Haokui Zhang, Wei Dong, Guoqing Wang, Peng Wang, Yang Yang, Hengtao Shen

A prevalent approach in Parameter-Efficient Fine-Tuning (PEFT) of pre-trained Vision Transformers (ViT) involves freezing the majority of the backbone parameters and solely learning low-rank adaptation weight matrices to accommodate downstream tasks. These low-rank matrices are commonly derived through the multiplication structure of down-projection and up-projection matrices, exemplified by methods such as LoRA and Adapter. In this study, we observe an approximate orthogonality among any two row or column vectors within any weight matrix of the backbone parameters; however, this property is absent in the vectors of the down/up-projection matrices. Approximate orthogonality implies a reduction in the upper bound of the model's generalization error, signifying that the model possesses enhanced generalization capability. If the fine-tuned down/up-projection matrices were to exhibit this same property as the pre-trained backbone matrices, could the generalization capability of fine-tuned ViTs be further augmented? To address this question, we propose an Approximately Orthogonal Fine-Tuning (AOFT) strategy for representing the low-rank weight matrices. This strategy employs a single learnable vector to generate a set of approximately orthogonal vectors, which form the down/up-projection matrices, thereby aligning the properties of these matrices with those of the backbone. Extensive experimental results demonstrate that our method achieves competitive performance across a range of downstream image classification tasks, confirming the efficacy of the enhanced generalization capability embedded in the down/up-projection matrices. Our code is available at link.

Goutham Vignesh, Harikrishnan P. M., Siddartha Reddy, Saisubramaniam Gopalakrishnan, Vishal Vaddina

Intelligent Document Processing (IDP) is critical for unlocking actionable insights from the vast volume of unstructured documents like invoices and medical reports, yet its promise is often unfulfilled as its implementation is typically hindered by significant technical barriers. Traditional IDP systems require deep expertise in programming, machine learning, and intricate model fine-tuning, creating a dependency on specialized data science teams. This effectively sidelines domain experts-the very individuals who possess the critical contextual understanding of the documents-thereby limiting the agility and accuracy of workflow automation. This paper introduces IDPFlow, a novel framework to unify a no-code, user-centric interface with a sophisticated, tool-augmented agentic architecture for end-to-end multimodal document processing, empowering experts such as business analysts and legal professionals to independently build and deploy sophisticated workflows without writing any code. IDPFlow is built upon a powerful agentic architecture, which intelligently utilize a versatile toolkit to execute a range of sophisticated IDP tasks. This toolkit enables a spectrum of high-precision IDP tasks such as multi-class document classification, Document visual question answering (Doc-VQA), key information extraction from text, tables, and checkboxes and long-document summarization. The core of IDPFlow is its dynamic agentic workflow, which redefines user interaction. Upon document upload, the agentic system instantly analyzes the content, classifying sub-documents and proactively suggesting a comprehensive data schema relevant to the use case, shifting the user's role from workflow builder to supervisor. This initial workflow is not static, it can be refined in real-time through simple, conversational instructions, enabling true business agility. Furthermore, the agentic intelligence extends to reusability, allowing existing workflows to be intelligently adapted for new, related tasks, dramatically reducing development time for subsequent use cases. For particularly complex tasks involving long or dense documents, the agentic system can leverage a specialized Multimodal Retrieval-Augmented Generation (MMRAG) pipeline to overcome the context window limitations of standard LLMs. This pipeline utilizes the ColPali model, which excels at generating unified multimodal embeddings, ensuring robust and accurate information retrieval from both textual content and embedded images or diagrams. To foster user trust and ensure verifiability, IDPFlow incorporates a grounded traceback citation mechanism that automatically highlights the precise document segments from which the agent derived its responses, making all outputs transparent and easily auditable. We highlight three key advantages of the framework: 1) Accessibility via an intuitive interface for domain experts; 2) Deep Adaptability and Reusability through dynamic agentic refinement and extensible tools; and 3) Trustworthiness rooted in a verifiable RAG pipeline and granular citation. The framework is projected to reduce end-to-end workflow creation time by 60-70% compared to traditional methods. Its unique combination of a no-code interface and a tool-augmented agentic architecture bridges the gap between technical complexity and domain expertise, accelerating the deployment of powerful, transparent, and scalable IDP solutions across industries.

Qiankun Li, Qiupu Chen, Huabao Chen, Feng He, Depeng Li 0001, Zhigang Zeng

Recent advances in video action recognition have achieved remarkable performance in coarse-grained macro-action classification by leveraging large-scale visual backbones and transformer architectures. However, extending these successes to fine-grained micro-action recognition remains a fundamental challenge due to the subtlety, brevity, and low motion intensity of micro-actions. In this paper, we propose a high-capacity framework for micro-action recognition, enhancing both representation learning and decision robustness. We scale to large-scale backbones using the VideoMAEv2 Giant model, enabling the extraction of finer spatial-temporal features. A Temporal-Spatial Connector (TSC) is introduced to dynamically highlight discriminative temporal frames and spatial regions, strengthening the model's focus on subtle motion cues critical for micro-action identification. To stabilize optimization and fully exploit the capacity of large models, we design a four-phase progressive training strategy, encompassing linear probing, full fine-tuning, connector-specific optimization, and classifier head refinement. Furthermore, we propose a novel ensemble decision mechanism that integrates Top-K predictions from diverse models via a Large Language Model (LLM), enhancing prediction consistency and robustness through multimodel consensus. Our method achieves an F1mean of 76.54% on the MA-52 dataset, ranking 3rd in the 2025 Micro-Action Analysis Grand Challenge and advancing the state of the art in fine-grained video understanding.

Robin-Nico Kampa, Fabian Deuser, Konrad Habel, Norbert Oswald

Plant phenotyping involves analyzing observable characteristics of plants to better understand their growth, health, and development. In the context of deep learning, this analysis is often approached through single-view classification or regression models. However, these methods often fail to capture all information required for accurate estimation of target phenotypic traits, which can adversely affect plant health assessment and harvest readiness prediction. To address this, the Growth Modelling (GroMo) Grand Challenge at ACM Multimedia 2025 provides a multi-view dataset featuring multiple plants and two tasks: Plant Age Prediction and Leaf Count Estimation. Each plant is photographed from multiple heights and angles, leading to significant overlap and redundancy in the captured information. To learn view-invariant embeddings, we incorporate 24 views, referred to as the selection vector, in a random selection. Our ViewSparsifier approach won both tasks. For further improvement and as a direction for future research, we also experimented with randomized view selection across all five height levels (120 views total), referred to as selection matrices.

Y. Hop Nguyen, Doan Anh Phan Huu, Trung Thai Tran, Nhat Nam Mai, Van Toi Giap, Thao Thi Phuong Dao, Trung-Nghia Le

We present a unified vision-language framework tailored for ENT endoscopy image analysis that simultaneously tackles three clinically-relevant tasks: image classification, image-to-image retrieval, and text-to-image retrieval. Unlike conventional CNN-based pipelines that struggle to capture cross-modal semantics, our approach leverages the CLIP ViT-B/16 backbone and enhances it through Low-Rank Adaptation, multi-level CLS token aggregation, and spherical feature interpolation. These components collectively enable efficient fine-tuning on limited medical data while improving representation diversity and semantic alignment across modalities. To bridge the gap between visual inputs and textual diagnostic context, we introduce class-specific natural language prompts that guide the image encoder through a joint training objective combining supervised classification with contrastive learning. We validated our framework through participation in the ACM MM'25 ENTRep Grand Challenge, achieving 95% accuracy and F1-score in classification, Recall@1 of 0.93 and 0.92 for image-to-image and text-to-image retrieval respectively, and MRR scores of 0.97 and 0.96. Ablation studies demonstrated the incremental benefits of each architectural component, validating the effectiveness of our design for robust multimodal medical understanding in low-resource clinical settings.