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.
论文检索
输入标题、作者或关键词，从 12,319 篇学术成果中精准定位
M2SFormer: Multi-Spectral and Multi-Scale Attention with Edge-Aware Difficulty Guidance for Image Forgery Localization
PDF ↗Image editing techniques have rapidly advanced, facilitating both innovative use cases and malicious manipulation of digital images. Deep learning-based methods have recently achieved high accuracy in pixel-level forgery localization, yet they frequently struggle with computational overhead and limited representation power, particularly for subtle or complex tampering. In this paper, we propose M2SFormer, a novel Transformer encoder-based framework designed to overcome these challenges. Unlike approaches that process spatial and frequency cues separately, M2SFormer unifies multi-frequency and multi-scale attentions in the skip connection, harnessing global context to better capture diverse forgery artifacts. Additionally, our framework addresses the loss of fine detail during upsampling by utilizing a global prior map--a curvature metric indicating the difficulty of forgery localization--which then guides a difficulty-guided attention module to preserve subtle manipulations more effectively. Extensive experiments on multiple benchmark datasets demonstrate that M2SFormer outperforms existing state-of-the-art models, offering superior generalization in detecting and localizing forgeries across unseen domains. Our M2SFormer code is available in Github Link.
PAN-sharpening aims to fuse high-resolution panchromatic (PAN) images with low-resolution multi-spectral (MS) images to generate high-resolution multi-spectral (HRMS) outputs. However, cross-modality misalignment---caused by sensor placement, acquisition timing, and resolution disparity---induces a fundamental challenge. Conventional deep learning methods assume perfect pixel-wise alignment and rely on per-pixel reconstruction losses, leading to spectral distortion, double edges, and blurring when misalignment is present. To address this, we propose PAN-Crafter, a modality-consistent alignment framework that explicitly mitigates the misalignment gap between PAN and MS modalities. At its core, Modality-Adaptive Reconstruction (MARs) enables a single network to jointly reconstruct HRMS and PAN images, leveraging PAN's high-frequency details as auxiliary self-supervision. Additionally, we introduce Cross-Modality Alignment-Aware Attention (CM3A), a novel mechanism that bidirectionally aligns MS texture to PAN structure and vice versa, enabling adaptive feature refinement across modalities. Extensive experiments on multiple benchmark datasets demonstrate that our PAN-Crafter outperforms the most recent state-of-the-art method in all metrics, even with 50.11xfaster inference time and 0.63xthe memory size. Furthermore, it demonstrates strong generalization performance on unseen satellite datasets, showing its robustness across different conditions.
Image composition has advanced significantly with large-scale pre-trained T2I diffusion models. Despite progress in same-domain composition, cross-domain composition remains under-explored. The main challenges are the stochastic nature of diffusion models and the style gap between input images, leading to failures and artifacts. Additionally, heavy reliance on text prompts limits practical applications. This paper presents the first cross-domain image composition method that does not require text prompts, allowing natural stylization and seamless compositions. Our method is efficient and robust, preserving the diffusion prior, as it involves minor steps for backward inversion and forward denoising without training the diffuser. Our method also uses a simple multilayer perceptron network to integrate CLIP features from foreground and background, manipulating diffusion with a local cross-attention strategy. It effectively preserves foreground content while enabling stable stylization without a pre-stylization network. Finally, we create a benchmark dataset with diverse contents and styles for fair evaluation, addressing the lack of testing datasets for cross-domain image composition. Our method outperforms state-of-the-art techniques in both qualitative and quantitative evaluations, significantly improving the LPIPS score by 30.5% and the CSD metric by 18.1%. We believe our method will advance future research and applications. Code and benchmark at https://github.com/sherlhw/AIComposer.
Generating body pose from head-mounted, egocentric inputs is essential for immersive VR/AR and assistive technologies, as it supports more natural interactions. However, the task is challenging due to limited visibility of body parts in first-person views and the sparseness of sensory data, with only a single device placed on the head. To address these challenges, we introduce Head2Body, a novel framework for body pose estimation that effectively combines IMU and visual data. First, we introduce a pre-trained IMU encoder, trained on over 1,700 hours of head-IMU data from wearable eyeglasses, to better capture detailed temporal motion cues given limited labeled egocentric pose data. For visual processing, we leverage large vision-language models (LVLMs) to segment body parts that appear sporadically in video frames to improve visual feature extraction. To better guide the pose generation process with sparse signals from only head-mounted devices, we incorporates a Vector Quantized Variational Autoencoder (VQ-VAE) to represent poses as discrete tokens, which capture high-frequency motion patterns and provide a more structured representation of body pose. Our experiments demonstrate the effectiveness of the proposed approach, yielding 6-13% gains over state-of-the-art baselines on four datasets: AMASS, KinPoly, GIMO, and EgoExo4D. By capturing subtle temporal dynamics and leveraging complementary sensory data, our approach advances accurate egocentric body pose estimation and sets a new benchmark for multi-modal, first-person motion tracking.
Recent advancements in large-scale video-language models have shown significant potential for real-time planning and detailed interactions. However, their high computational demands and the scarcity of annotated datasets limit their practicality for academic researchers. In this work, we introduce VideoLLaMB, a novel and efficient framework for long video understanding that leverages recurrent memory bridges and temporal memory tokens to enable seamless encoding of entire video sequences with preserved semantic continuity. Central to our approach is a SceneTiling algorithm that segments videos into coherent semantic units, facilitating robust understanding across tasks without requiring additional training. VideoLLaMB achieves state-of-the-art performance, surpassing existing models by 4.2 points on four VideoQA benchmarks and by 2.06 points on egocentric planning tasks. Notably, it maintains strong performance under extreme video length scaling (up to 8x) and excels at fine-grained frame retrieval on our proposed Needle in a Video Haystack (NIAVH) benchmark. With linear GPU memory scaling, VideoLLaMB processes up to 320 frames using a single Nvidia A100 GPU, despite being trained on only 16 frames--offering an unprecedented balance of accuracy, scalability, and cost-effectiveness. This makes it highly accessible and practical for the academic community.
Mathematical problems in real-world scenarios are often presented in a purely vision-form, where textual problem statement and accompanying math figures, e.g., geometry figures and functional graphs, are integrated into a single image. This vision-form problem-solving task requires precise comprehension and reasoning on both textual and graphical elements in the images, posing significant challenge to current Multimodal Large Language Models (MLLMs), which process text and math figures in isolation. In this work, we propose VisionMath, the first exploration for vision-form mathematical problem-solving model, which employs a three-stage progressive multimodal reasoning alignment strategy to systematically enhance task-specific capabilities. Building upon a LLM proficient in unimodal mathematical reasoning, VisionMath first establishes foundational OCR capabilities through capturing rendered mathematical problem images. Subsequently, the model develops comprehensive understanding of figure structures and properties via learning from figure descriptions and mathematical educational videos. Finally, the model's reasoning capacity is activated using carefully constructed visual-form problem-solving datasets VisionMath-IT with chain-of-thought annotations. For comprehensive evaluation, we construct multilingual benchmarks covering diverse problem types, including geometry, algebra, function problems in both English and Chinese. Our model weights, data and code will be made available at https://github.com/mengqiDyangge/VisionMath.
Vision Language Models (VLMs) have lead to major improvements in multimodal reasoning, yet they still struggle to understand user-specific concepts. Existing personalization methods address this limitation butheavily rely on training procedures, that can be either costly or unpleasant to individual users.We depart from existing work, and for the first time explore the training-free setting in the context of personalization. We propose a novel method, Retrieval and Reasoning for Personalization (R2P), leveraging internal knowledge of VLMs. First, we leverage VLMs to extract the concept fingerprint, i.e., key attributes uniquely defining the concept within its semantic class. When a query arrives, the most similar fingerprints are retrieved and scored via chain of thought reasoning. To reduce the risk of hallucinations, the scores are validated through cross-modal verification at the attribute level:in case of a discrepancy between the scores, R2P refines the concept association viapairwise multimodal matching, where the retrieved fingerprints and their images aredirectly compared with the query.We validate R2P on two publicly available benchmarks and a newly introduced dataset, Personal Concepts with Visual Ambiguity (PerVA), for concept identification highlighting challenges in visual ambiguity. R2P consistently outperforms state-of-the-art approaches on various downstream tasks across all benchmarks.
Model-heterogeneous federated learning (MHFL) is a challenging FL paradigm designed to allow FL clients to train structurally heterogeneous models under the coordination of an FL server. Existing MHFL methods face significant limitations when it comes to transferring global knowledge to clients as a result of sharing only partial homogeneous model parameters or calculating distance loss, leading to inferior model generalization. To bridge this gap, we propose a novel model-heterogeneous Federated learning method with Representation Angle Learning (FedRAL). It consists of three innovative designs: (1) We first introduce representation angle learning into MHFL. Specifically, we embed a homogeneous square matrix into the local heterogeneous model of each client, which learns the angle information of local representations. These homogeneous representation angle square matrices are aggregated on the server to fuse representation angle knowledge shared by clients for enhancing the generalization of local representations. (2) As different clients might have heterogeneous system resources, we propose an adaptive diagonal sparsification strategy to reduce the numbers of the parameters of representation angle square matrices uploaded to the server, to improve FL communication efficiency. (3) To enable the effective fusion of sparsified homogeneous local representation angle square matrices, we design an element-wise weighted aggregation approach. Experiments on 4 benchmark datasets under 2 types of non-IID divisions over 6 state-of-the-art baselines demonstrate that FedRAL achieves the best performance. It improves test accuracy, communication efficiency and computational efficiency by up to 5.03%, 12.43x and 6.49x, respectively.
Traditional alignment methods for Large Vision and Language Models (LVLMs) primarily rely on human-curated preference data. Human-generated preference data is costly; machine-generated preference data is limited in quality; and self-supervised preference data often introduces hallucinations. To overcome these limitations, we propose a novel Panel-of-Peers learning framework inspired by collaborative learning among humans. This approach leverages a panel of LVLMs, each evaluating and learning from their collective outputs through an iterative self-improvement process. By simulating a peer review system, our models generate, assess, and refine outputs in response to a curated set of prompts, mimicking a classroom learning environment. We demonstrate that this methodology enhances model performance without requiring extensive human-labeled datasets. Our experiments show significant improvement across multiple benchmarks, demonstrating the potential of peer evaluations as a scalable alternative to self-supervised alignment. Notably, we show that Panel-of-Peers increases the average score on fifteen benchmarks from 48% to 57%.
ROADWork: A Dataset and Benchmark for Learning to Recognize, Observe, Analyze and Drive Through Work Zones
PDF ↗Perceiving and autonomously navigating through work zones is a challenging and under-explored problem. Open datasets for this long-tailed scenario are scarce. We propose the ROADWork dataset to learn to recognize, observe, analyze, and drive through work zones. State-of-the-art foundation models fail when applied to work zones. Fine-tuning models on our dataset significantly improves perception and navigation in work zones. With ROADWork, we discover new work zone images with higher precision (+32.5%) at a much higher rate (12.8x) around the world. Open-vocabulary methods fail too, whereas fine-tuned detectors improve performance (+32.2 AP).Vision-Language Models (VLMs) struggle to describe work zones, but fine-tuning substantially improves performance (+36.7 SPICE). Beyond fine-tuning, we show the value of simple techniques. Video label propagation provides additional gains (+2.6 AP) for instance segmentation. While reading work zone signs, composing a detector and text spotter via crop-scaling improves performance (+14.2% 1-NED). Composing work zone detections to provide context further reduces hallucinations (+3.9 SPICE) in VLMs. We predict navigational goals and compute drivable paths from work zone videos. Incorporating road work semantics ensures 53.6% goals have angular error (AE) < 0.5 (+9.9%) and 75.3% pathways have AE < 0.5 (+8.1%).
Advances in Earth observation (EO) foundation models have unlocked the potential of big satellite data to learn generic representations from space, benefiting a wide range of downstream applications crucial to our planet. However, most existing efforts remain limited to fixed spectral sensors, focus solely on the Earth's surface, and overlook valuable metadata beyond imagery. In this work, we take a step towards next-generation EO foundation models with three key components: 1) Copernicus-Pretrain, a massive-scale pretraining dataset that integrates 18.7M aligned images from all major Copernicus Sentinel missions, spanning from the Earth's surface to its atmosphere; 2) Copernicus-FM, a unified foundation model capable of processing any spectral or non-spectral sensor modality using extended dynamic hypernetworks and flexible metadata encoding; and 3) Copernicus-Bench, a systematic evaluation benchmark with 15 hierarchical downstream tasks ranging from preprocessing to specialized applications for each Sentinel mission. Our dataset, model, and benchmark greatly improve the scalability, versatility, and multimodal adaptability of EO foundation models, while also creating new opportunities to connect EO, weather, and climate research. Codes at https://github.com/zhu-xlab/Copernicus-FM.
The movement for Sovereign AI is accelerating. Meeting its promise requires vertically integrated AI stacks -spanning data, models, and reasoning systems- that remain sovereign while adhering to shared scientific principles around which global research communities can coalesce. This talk presents BharatGen as a sovereign-yet-shared effort to make AI work for all: creation of datasets, benchmarks, and models that natively support Indian languages, dialects, and code mixing across text, speech, and vision; data pipelines grounded in local realities; and frugal methods that reduce cost and lower barriers. We outline our journey to date across language infrastructure, efficient training and distillation, and early sector pilots. The R&D deep dive will draw from some of our recent work on cross-lingual knowledge distillation for low-resource languages, tokenization/phonetic design for code-mix robustness, or trustworthy document AI with visual grounding focusing on robustness under dialect/code-mix shift, and latency/cost trade-offs. We hope to inspire other Sovereign-AI efforts, especially in the low-resource ecosystems of the Global South and close by inviting international collaborations toward principled research to build people-serving AI.
We present MedAI Hub, an integrated multimodal medical platform designed to bridge clinical practice and research by transforming patient-doctor interactions into structured scientific data. This platform supports comprehensive management of multimodal medical records-including clinical notes, medical images, and patient-reported outcomes-while implementing privacy-preserving data sharing mechanisms. Building upon this infrastructure, we introduce two novel AI-driven modules:(1) ITERATE (Image-Text Enhancement, Retrieval, and Alignment): An evolutionary algorithm inspired by Visual Genome that optimizes medical image-text alignment through iterative cross-modal refinement. Leveraging LLM-guided ''DNA evolution'' and multimodal feedback, ITERATE enhances ultrasound image quality for diagnostic tasks, achieving 3.5-7% accuracy gains on ScienceQA and ARC-Easy benchmarks.(2) MedQuery: A graph-driven literature retrieval system that constructs multimodal knowledge graphs from medical literature (text, figures, tables). By aligning PubMed documents with complex clinical queries through semantic relationship modeling, it achieves >90% answer quality win rates and 13-36% accuracy improvements on PubMedQA and MedInquiry datasets. MedAI Hub demonstrates that synergistic integration of clinical data platforms with evolutionary vision-language optimization and multimodal knowledge graphs significantly advances medical AI capabilities, enabling more accurate diagnostics and research insights. The platform and algorithms are publicly available to accelerate innovation in medical AI.
This workshop addresses next-generation methods in multimedia research, with a focus on content generation, quality assessment, and dataset development. These three areas are foundational for advancing multimedia technologies and applications. Emerging approaches in multimedia content generation, powered by generative AI and multimodal learning, are reshaping domains such as entertainment, advertising, education, and healthcare. At the same time, robust quality assessment is essential to ensure that generated content achieves high standards of perceptual fidelity, semantic consistency, and user satisfaction, thereby determining the real-world impact of multimedia systems. Datasets remain indispensable for training and evaluating algorithms, and innovative strategies in dataset construction-ranging from augmentation and annotation to addressing issues of bias and small-sample imbalance-are driving the development of more reliable and ethical multimedia applications. By convening leading researchers and practitioners, this workshop provides a platform to explore state-of-the-art methods, share best practices, and discuss open challenges in next-generation multimedia research. The goal is to foster interdisciplinary collaboration and inspire innovative solutions that advance the creation, evaluation, and application of multimedia content, setting new benchmarks for the field and shaping the future of multimedia technologies.
The proliferation of generative models, particularly Generative Adversarial Networks (GANs) and Diffusion Models, has reshaped multimedia content creation. Alongside creative and commercial opportunities, they have introduced unprecedented risks through the production of highly realistic synthetic content, or deepfakes. These artifacts challenge visual and auditory trust, with major implications for media, security, politics, and law. This workshop provides a forum to examine deepfake technology from forensic, technical, legal, and social perspectives. It will bring together experts to advance robust and explainable detection methods, define benchmarking practices, and address ethical and regulatory frameworks. Topics include detection and attribution, adversarial countermeasures, multimodal analysis, model traceability, legal admissibility of synthetic content, as well as real-world deployment challenges and dataset creation. Further information about the workshop is available at https://iplab.dmi.unict.it/mfs/acm-dff-ws-2025/
The Event-Enriched Image Analysis (EVENTA) Grand Challenge, hosted at ACM Multimedia 2025, introduces the first large-scale benchmark for event-level multimodal understanding. Traditional captioning and retrieval tasks largely focus on surface-level recognition of people, objects, and scenes, often overlooking the contextual and semantic dimensions that define real-world events. EVENTA addresses this gap by integrating contextual, temporal, and semantic information to capture the who, when, where, what, and why behind an image. Built upon the OpenEvents V1 dataset, the challenge features two tracks: Event-Enriched Image Retrieval and Captioning, and Event-Based Image Retrieval. A total of 45 teams from six countries participated, with evaluation conducted through Public and Private Test phases to ensure fairness and reproducibility. The top three teams were invited to present their solutions at ACM Multimedia 2025. EVENTA establishes a foundation for context-aware, narrative-driven multimedia AI, with applications in journalism, media analysis, cultural archiving, and accessibility. Further details about the challenge are available at the official homepage: https://ltnghia.github.io/eventa/eventa-2025.
Micro-Actions (MAs) are a crucial form of non-verbal communication in social interactions, with promising applications in human emotion analysis. Although the topic has attracted considerable research interest, progress has been hindered by the lack of publicly available benchmark datasets. To address this gap, the Micro-Action Analysis Grand Challenge (MAC) is organized annually. This paper presents an overview of the 2nd Micro-Action Analysis Grand Challenge, held in conjunction with ACM Multimedia 2025. We provide a comprehensive summary of the challenge, including its dataset, evaluation protocol, results, and discussion. The top-ranked solutions are highlighted to offer valuable insights for researchers, and potential future directions are outlined to guide ongoing developments in this area. The goal of this grand challenge is to foster innovative research in micro-action analysis and advance research in the human-centric action understanding community.
Understanding plant growth dynamics is a critical component of modern agricultural research, with applications in yield prediction, phenotyping, and sustainable crop management. Despite recent advances in computer vision and deep learning, progress in plant growth modeling has been constrained by the lack of publicly available, high-resolution, multiview, and temporally rich datasets. To address this gap, we introduce Growth Modelling GroMo25, the first international challenge on plant growth modeling using multiview imagery. In this challenge, we propose a dataset that comprises high-resolution images of four crops: wheat, mustard, radish, and okra, captured at consistent time intervals from multiple camera viewpoints under controlled environmental conditions. The challenge focuses on two key tasks: (1) plant age prediction and (2) leaf count estimation, both requiring models to use spatial and temporal plant features. GroMo25 attracted participation from multiple teams worldwide, encouraging benchmarking and innovation in vision-based plant phenotyping. The GitHub repository is publicly available at https://github.com/mriglab/GroMo-Plant-Growth-Modeling-with-Multiview-Images.
Classifying anatomical regions in endoscopic ENT (ear, nose, and throat) images is challenging due to strong inter-class similarities, bilateral symmetry, and the scarcity of annotated datasets. To overcome these issues, we present HyMoENet. This novel hybrid deep learning architecture combines convolutional neural networks (CNNs) for localized feature extraction with Vision Transformers to represent global context. Furthermore, it leverages a sparse Mixture-of-Experts (MoE) technique to improve multi-perspective specialization. Our architecture makes use of parallel CNN-Transformer encoders, which are incorporated into a dynamic MoE layer that adaptively routes representations to the most appropriate experts. Concurrently, a semantic-preserving skip connection preserves global coherence. When tested on a clinically annotated ENT endoscopy dataset from Thong Nhat Hospital in Vietnam, HyMoENet outperformed both single-stream and conventional hybrid models with an accuracy of 97.50. These results demonstrate that integrating local-global representation learning and expert modularization enhances classification accuracy for anatomically similar structures. HyMoENet sets a new benchmark for automated ENT image processing, setting the groundwork for intelligent diagnostic systems in clinical endoscopy.