Caption
-
Learning to Think Like a Cartoon Captionist: Incongruity-Resolution Supervision for Multimodal Humor Understanding
2026-04-16 16:41:13Hatice Merve Vural, Doga Kukul, Ege Erdem Ozlu, Demir Ekin Arikan, Bob Mankoff, Erkut Erdem, Aykut ErdemAbstract
Humor is one of the few cognitive tasks where getting the reasoning right matters as much as getting the answer right. While recent work evaluates humor understanding on benchmarks such as the New Yorker Cartoon Caption Contest (NYCC), it largely treats it as black-box prediction, overlooking the structured reasoning processes underlying humor comprehension. We introduce IRS (Incongruity-Resolution Supervision), a framework that decomposes humor understanding into three components: incongruity modeling, which identifies mismatches in the visual scene; resolution modeling, which constructs coherent reinterpretations of these mismatches; and preference alignment, which evaluates candidate interpretations under human judgments. Grounded in incongruity-resolution theory and expert captionist practice, IRS supervises intermediate reasoning process through structured traces that make the path from visual perception to humorous interpretation explicit and learnable. Across 7B, 32B, and 72B models on NYCC, IRS outperforms strong open and closed multimodal baselines across caption matching and ranking tasks, with our largest model approaching expert-level performance on ranking. Zero-shot transfer to external benchmarks shows that IRS learns generalizable reasoning patterns. Our results suggest that supervising reasoning structure, rather than scale alone, is key for reasoning-centric tasks.
Abstract (translated)
URL
https://arxiv.org/abs/2604.15210
PDF
-
MetaDent: Labeling Clinical Images for Vision-Language Models in Dentistry
2026-04-16 10:56:54Meng-Xun Li, Wen-Hui Deng, Zhi-Xing Wu, Chun-Xiao Jin, Jia-Min Wu, Yue Han, James Kit Hon Tsoi, Gui-Song Xia, Cui HuangAbstract
Vision-Language Models (VLMs) have demonstrated significant potential in medical image analysis, yet their application in intraoral photography remains largely underexplored due to the lack of fine-grained, annotated datasets and comprehensive benchmarks. To address this, we present MetaDent, a comprehensive resource that includes (1) a novel and large-scale dentistry image dataset collected from clinical, public, and web sources; (2) a semi-structured annotation framework designed to capture the hierarchical and clinically nuanced nature of dental photography; and (3) comprehensive benchmark suites for evaluating state-of-the-art VLMs on clinical image understanding. Our labeling approach combines a high-level image summary with point-by-point, free-text descriptions of abnormalities. This method enables rich, scalable, and task-agnostic representations. We curated 60,669 dental images from diverse sources and annotated a representative subset of 2,588 images using this meta-labeling scheme. Leveraging Large Language Models (LLMs), we derive standardized benchmarks: approximately 15K Visual Question Answering (VQA) pairs and an 18-class multi-label classification dataset, which we validated with human review and error analysis to justify that the LLM-driven transition reliably preserves fidelity and semantic accuracy. We then evaluate state-of-the-art VLMs across VQA, classification, and image captioning tasks. Quantitative results reveal that even the most advanced models struggle with a fine-grained understanding of intraoral scenes, achieving moderate accuracy and producing inconsistent or incomplete descriptions in image captioning. We publicly release our dataset, annotations, and tools to foster reproducible research and accelerate the development of vision-language systems for dental applications.
Abstract (translated)
URL
https://arxiv.org/abs/2604.14866
PDF
-
Chain-of-Glimpse: Search-Guided Progressive Object-Grounded Reasoning for Video Understanding
2026-04-16 06:50:20Zhixuan Wu, Quanxing Zha, Teng Wang, Genbao Xu, Wenyuan Gu, Wei Rao, Nan Ma, Bo Cheng, Soujanya PoriaAbstract
Video understanding requires identifying and reasoning over semantically discriminative visual objects across frames, yet existing object-agnostic solutions struggle to effectively handle substantial object variations over time. To address this, we introduce Chain-of-Glimpse, a search-guided progressive object-grounded reasoning framework that explicitly anchors each reasoning step to specific visual evidence regions, enabling compositional and multi-step decision-making. Formally, Chain-of-Glimpse formulates video reasoning as a step-by-step process that incrementally builds spatially grounded traces around task-relevant visual objects, thereby mitigating over-reliance on saliency-driven cues. Specifically, Chain-of-Glimpse features a search-guided controller, optimized via reinforcement learning with a format reward that significantly incentivizes grounding capability, to iteratively ground visual evidence regions and form reliable reasoning trajectories, yielding accurate and interpretable multi-step decisions. Extensive evaluations on both in domain NExTQA and out-of-domain Video-Holmes, CG-Bench Reasoning, and VRBench benchmarks demonstrate consistent performance gains, robustness and generalization of Chain-of-Glimpse across diverse video reasoning tasks.
Abstract (translated)
URL
https://arxiv.org/abs/2604.14692
PDF
-
SatBLIP: Context Understanding and Feature Identification from Satellite Imagery with Vision-Language Learning
2026-04-15 19:43:20Xue Wu, Shengting Cao, Jiaqi GongAbstract
Rural environmental risks are shaped by place-based conditions (e.g., housing quality, road access, land-surface patterns), yet standard vulnerability indices are coarse and provide limited insight into risk contexts. We propose SatBLIP, a satellite-specific vision-language framework for rural context understanding and feature identification that predicts county-level Social Vulnerability Index (SVI). SatBLIP addresses limitations of prior remote sensing pipelines-handcrafted features, manual virtual audits, and natural-image-trained VLMs-by coupling contrastive image-text alignment with bootstrapped captioning tailored to satellite semantics. We use GPT-4o to generate structured descriptions of satellite tiles (roof type/condition, house size, yard attributes, greenery, and road context), then fine-tune a satellite-adapted BLIP model to generate captions for unseen images. Captions are encoded with CLIP and fused with LLM-derived embeddings via attention for SVI estimation under spatial aggregation. Using SHAP, we identify salient attributes (e.g., roof form/condition, street width, vegetation, cars/open space) that consistently drive robust predictions, enabling interpretable mapping of rural risk environments.
Abstract (translated)
URL
https://arxiv.org/abs/2604.14373
PDF
-
One Token per Highly Selective Frame: Towards Extreme Compression for Long Video Understanding
2026-04-15 17:59:52Zheyu Zhang, Ziqi Pang, Shixing Chen, Xiang Hao, Vimal Bhat, Yu-Xiong WangAbstract
Long video understanding is inherently challenging for vision-language models (VLMs) because of the extensive number of frames. With each video frame typically expanding into tens or hundreds of tokens, the limited context length of large language models (LLMs) forces the VLMs to perceive the frames sparsely and lose temporal information. To address this, we explore extreme video token compression towards \emph{one token per frame} at the final LLM layer. Our key insight is that heuristic-based compression, widely adopted by previous methods, is prone to information loss, and this necessitates supervising LLM layers into \emph{learnable} and \emph{progressive} modules for \emph{token-level compression} (LP-Comp). Such compression enables our VLM to digest 2x-4x more frames with improved performance. To further increase the token efficiency, we investigate \emph{frame-level compression}, which selects the frames most relevant to the queries via the internal attention scores of the LLM layers, named \emph{question-conditioned compression} (QC-Comp). As a notable distinction from previous studies, we mitigate the position bias of LLM attention in long contexts, \emph{i.e.}, the over-concentration on the beginning and end of a sequence, by splitting long videos into short segments and employing local attention. Collectively, our combined \emph{token-level} and \emph{frame-level} leads to an e\textbf{x}treme compression model for long video understanding, named \textbf{\name}, achieving a significantly larger compression ratio and enabling denser frame sampling. Our \name is finetuned from VideoChat-Flash with a data-efficient \emph{supervised compression tuning} stage that only requires 2.5\% of the supervised fine-tuning data, yet boosts the accuracy from 42.9\% to 46.2\% on LVBench and enhances multiple other long video benchmarks.
Abstract (translated)
URL
https://arxiv.org/abs/2604.14149
PDF
-
Training-Free Semantic Multi-Object Tracking with Vision-Language Models
2026-04-15 16:44:57Laurence Bonat, Francesco Tonini, Elisa Ricci, Lorenzo VaqueroAbstract
Semantic Multi-Object Tracking (SMOT) extends multi-object tracking with semantic outputs such as video summaries, instance-level captions, and interaction labels, aiming to move from trajectories to human-interpretable descriptions of dynamic scenes. Existing SMOT systems are trained end-to-end, coupling progress to expensive supervision, limiting the ability to rapidly adapt to new foundation models and new interactions. We propose TF-SMOT, a training-free SMOT pipeline that composes pretrained components for detection, mask-based tracking, and video-language generation. TF-SMOT combines D-FINE and the promptable SAM2 segmentation tracker to produce temporally consistent tracklets, uses contour grounding to generate video summaries and instance captions with InternVideo2.5, and aligns extracted interaction predicates to BenSMOT WordNet synsets via gloss-based semantic retrieval with LLM disambiguation. On BenSMOT, TF-SMOT achieves state-of-the-art tracking performance within the SMOT setting and improves summary and caption quality compared to prior art. Interaction recognition, however, remains challenging under strict exact-match evaluation on the fine-grained and long-tailed WordNet label space; our analysis and ablations indicate that semantic overlap and label granularity substantially affect measured performance.
Abstract (translated)
URL
https://arxiv.org/abs/2604.14074
PDF
-
Towards Fine-grained Temporal Perception: Post-Training Large Audio-Language Models with Audio-Side Time Prompt
2026-04-15 10:50:29Yanfeng Shi, Pengfei Cai, Jun Liu, Qing Gu, Nan Jiang, Lirong Dai, Ian McLoughlin, Yan SongAbstract
Large Audio-Language Models (LALMs) enable general audio understanding and demonstrate remarkable performance across various audio tasks. However, these models still face challenges in temporal perception (e.g., inferring event onset and offset), leading to limited utility in fine-grained scenarios. To address this issue, we propose Audio-Side Time Prompt and leverage Reinforcement Learning (RL) to develop the TimePro-RL framework for fine-grained temporal perception. Specifically, we encode timestamps as embeddings and interleave them within the audio feature sequence as temporal coordinates to prompt the model. Furthermore, we introduce RL following Supervised Fine-Tuning (SFT) to directly optimize temporal alignment performance. Experiments demonstrate that TimePro-RL achieves significant performance gains across a range of audio temporal tasks, such as audio grounding, sound event detection, and dense audio captioning, validating its robust effectiveness.
Abstract (translated)
URL
https://arxiv.org/abs/2604.13715
PDF
-
Enhanced Text-to-Image Generation by Fine-grained Multimodal Reasoning
2026-04-15 05:24:29Yongjin Kim, Yoonjin Oh, Yerin Kim, Hyomin Kim, Jeeyoung Yun, Yujung Heo, Minjun Kim, Sungwoong KimAbstract
With the rapid progress of Multimodal Large Language Models (MLLMs), unified MLLMs that jointly perform image understanding and generation have advanced significantly. However, despite the inherent reasoning capabilities of unified MLLMs for self-reflection and self-refinement, their use in text-to-image generation remains largely underexplored. Meanwhile, existing multimodal reasoning-based image generation methods mostly rely on holistic image-text alignment judgments, without fine-grained reflection and refinement of detailed prompt attributes, leading to limited fine-grained control. Therefore, we propose Fine-grained Multimodal Reasoning (FiMR), a framework that leverages decomposed visual question answering (VQA) to break down an input prompt into minimal semantic units-such as entities and attributes-and verify each unit via VQA to generate explicit, fine-grained feedback. Based on this feedback, FiMR then applies targeted, localized refinements. This fine-grained self-reasoning and self-refinement enable MLLMs to achieve more precise improvements in image-prompt alignment and overall generation quality at test time. Extensive experiments demonstrate that FiMR consistently outperforms image generation baselines, including reasoning-based methods, particularly on compositional text-to-image benchmarks.
Abstract (translated)
URL
https://arxiv.org/abs/2604.13491
PDF
-
A Sanity Check on Composed Image Retrieval
2026-04-14 15:52:22Yikun Liu, Jiangchao Yao, Weidi Xie, Yanfeng WangAbstract
Composed Image Retrieval (CIR) aims to retrieve a target image based on a query composed of a reference image, and a relative caption that specifies the desired modification. Despite the rapid development of CIR models, their performance is not well characterized by existing benchmarks, which inherently contain indeterminate queries degrading the evaluation (i.e., multiple candidate images, rather than solely the target image, meet the query criteria), and have not considered their effectiveness in the context of the multi-round system. Motivated by this, we consider improving the evaluation procedure from two aspects: 1) we introduce FISD, a Fully-Informed Semantically-Diverse benchmark, which employs generative models to precisely control the variables of reference-target image pairs, enabling a more accurate evaluation of CIR methods across six dimensions, without query ambiguity; 2) we propose an automatic multi-round agentic evaluation framework to probe the potential of the existing models in the interactive scenarios. By observing how models adapt and refine their choices over successive rounds of queries, this framework provides a more realistic appraisal of their efficacy in practical applications. Extensive experiments and comparisons prove the value of our novel evaluation on typical CIR methods.
Abstract (translated)
URL
https://arxiv.org/abs/2604.12904
PDF
-
Challenging Vision-Language Models with Physically Deployable Multimodal Semantic Lighting Attacks
2026-04-14 14:52:15Yingying Zhao, Chengyin Hu, Qike Zhang, Xin Li, Xin Wang, Yiwei Wei, Jiujiang Guo, Jiahuan Long, Tingsong Jiang, Wen YaoAbstract
Vision-Language Models (VLMs) have shown remarkable performance, yet their security remains insufficiently understood. Existing adversarial studies focus almost exclusively on the digital setting, leaving physical-world threats largely unexplored. As VLMs are increasingly deployed in real environments, this gap becomes critical, since adversarial perturbations must be physically realizable. Despite this practical relevance, physical attacks against VLMs have not been systematically studied. Such attacks may induce recognition failures and further disrupt multimodal reasoning, leading to severe semantic misinterpretation in downstream tasks. Therefore, investigating physical attacks on VLMs is essential for assessing their real-world security risks. To address this gap, we propose Multimodal Semantic Lighting Attacks (MSLA), the first physically deployable adversarial attack framework against VLMs. MSLA uses controllable adversarial lighting to disrupt multimodal semantic understanding in real scenes, attacking semantic alignment rather than only task-specific outputs. Consequently, it degrades zero-shot classification performance of mainstream CLIP variants while inducing severe semantic hallucinations in advanced VLMs such as LLaVA and BLIP across image captioning and visual question answering (VQA). Extensive experiments in both digital and physical domains demonstrate that MSLA is effective, transferable, and practically realizable. Our findings provide the first evidence that VLMs are highly vulnerable to physically deployable semantic attacks, exposing a previously overlooked robustness gap and underscoring the urgent need for physical-world robustness evaluation of VLMs.
Abstract (translated)
URL
https://arxiv.org/abs/2604.12833
PDF
-
A Multi-Agent Feedback System for Detecting and Describing News Events in Satellite Imagery
2026-04-14 14:12:19Madeline Anderson, Mikhail Klassen, Ash Hoover, Kerri CahoyAbstract
Changes in satellite imagery often occur over multiple time steps. Despite the emergence of bi-temporal change captioning datasets, there is a lack of multi-temporal event captioning datasets (at least two images per sequence) in remote sensing. This gap exists because (1) searching for visible events in satellite imagery and (2) labeling multi-temporal sequences require significant time and labor. To address these challenges, we present SkyScraper, an iterative multi-agent workflow that geocodes news articles and synthesizes captions for corresponding satellite image sequences. Our experiments show that SkyScraper successfully finds 5x more events than traditional geocoding methods, demonstrating that agentic feedback is an effective strategy for surfacing new multi-temporal events in satellite imagery. We apply our framework to a large database of global news articles, curating a new multi-temporal captioning dataset with 5,000 sequences. By automatically identifying imagery related to news events, our work also supports journalism and reporting efforts.
Abstract (translated)
URL
https://arxiv.org/abs/2604.12772
PDF
-
PromptEcho: Annotation-Free Reward from Vision-Language Models for Text-to-Image Reinforcement Learning
2026-04-14 12:21:15Jinlong Liu, Wanggui He, Peng Zhang, Mushui Liu, Hao Jiang, Pipei HuangAbstract
Reinforcement learning (RL) can improve the prompt following capability of text-to-image (T2I) models, yet obtaining high-quality reward signals remains challenging: CLIP Score is too coarse-grained, while VLM-based reward models (e.g., RewardDance) require costly human-annotated preference data and additional fine-tuning. We propose PromptEcho, a reward construction method that requires \emph{no} annotation and \emph{no} reward model training. Given a generated image and a guiding query, PromptEcho computes the token-level cross-entropy loss of a frozen VLM with the original prompt as the label, directly extracting the image-text alignment knowledge encoded during VLM pretraining. The reward is deterministic, computationally efficient, and improves automatically as stronger open-source VLMs become available. For evaluation, we develop DenseAlignBench, a benchmark of concept-rich dense captions for rigorously testing prompt following capability. Experimental results on two state-of-the-art T2I models (Z-Image and QwenImage-2512) demonstrate that PromptEcho achieves substantial improvements on DenseAlignBench (+26.8pp / +16.2pp net win rate), along with consistent gains on GenEval, DPG-Bench, and TIIFBench without any task-specific training. Ablation studies confirm that PromptEcho comprehensively outperforms inference-based scoring with the same VLM, and that reward quality scales with VLM size. We will open-source the trained models and the DenseAlignBench.
Abstract (translated)
URL
https://arxiv.org/abs/2604.12652
PDF
-
Relaxing Anchor-Frame Dominance for Mitigating Hallucinations in Video Large Language Models
2026-04-14 11:05:42Zijian Liu, Sihan Cao, Pengcheng Zheng, Kuien Liu, Caiyan Qin, Xiaolin Qin, Jiwei Wei, Chaoning ZhangAbstract
Recent Video Large Language Models (Video-LLMs) have demonstrated strong capability in video understanding, yet they still suffer from hallucinations. Existing mitigation methods typically rely on training, input modification, auxiliary guidance, or additional decoding procedures, while largely overlooking a more fundamental challenge. During generation, Video-LLMs tend to over-rely on a limited portion of temporal evidence, leading to temporally imbalanced evidence aggregation across the video. To address this issue, we investigate a decoder-side phenomenon in which the model exhibits a temporally imbalanced concentration pattern. We term the frame with the highest aggregated frame-level attention mass the anchor frame. We find that this bias is largely independent of the input video and instead appears to reflect a persistent, model-specific structural or positional bias, whose over-dominance is closely associated with hallucination-prone generation. Motivated by this insight, we propose Decoder-side Temporal Rebalancing (DTR), a training-free, layer-selective inference method that rebalances temporal evidence allocation in middle-to-late decoder layers without altering visual encoding or requiring auxiliary models. DTR adaptively calibrates decoder-side visual attention to alleviate temporally imbalanced concentration and encourage under-attended frames to contribute more effectively to response generation. In this way, DTR guides the decoder to ground its outputs in temporally broader and more balanced video evidence. Extensive experiments on hallucination and video understanding benchmarks show that DTR consistently improves hallucination robustness across diverse Video-LLM families, while preserving competitive video understanding performance and high inference efficiency.
Abstract (translated)
URL
https://arxiv.org/abs/2604.12582
PDF
-
ReflectCAP: Detailed Image Captioning with Reflective Memory
2026-04-14 06:47:47Kyungmin Min, Minbeom Kim, Kang-il Lee, Seunghyun Yoon, Kyomin JungAbstract
Detailed image captioning demands both factual grounding and fine-grained coverage, yet existing methods have struggled to achieve them simultaneously. We address this tension with Reflective Note-Guided Captioning (ReflectCAP), where a multi-agent pipeline analyzes what the target large vision-language model (LVLM) consistently hallucinates and what it systematically overlooks, distilling these patterns into reusable guidelines called Structured Reflection Notes. At inference time, these notes steer the captioning model along both axes -- what to avoid and what to attend to -- yielding detailed captions that jointly improve factuality and coverage. Applying this method to 8 LVLMs spanning the GPT-4.1 family, Qwen series, and InternVL variants, ReflectCAP reaches the Pareto frontier of the trade-off between factuality and coverage, and delivers substantial gains on CapArena-Auto, where generated captions are judged head-to-head against strong reference models. Moreover, ReflectCAP offers a more favorable trade-off between caption quality and compute cost than model scaling or existing multi-agent pipelines, which incur 21--36\% greater overhead. This makes high-quality detailed captioning viable under real-world cost and latency constraints.
Abstract (translated)
URL
https://arxiv.org/abs/2604.12357
PDF
-
Unlocking the Potential of Grounding DINO in Videos: Parameter-Efficient Adaptation for Limited-Data Spatial-Temporal Localization
2026-04-14 06:32:24Zanyi Wang, Fan Li, Dengyang Jiang, Liuzhuozheng Li, Yunhua Zhong, Guang Dai, Mengmeng WangAbstract
Spatio-temporal video grounding (STVG) aims to localize queried objects within dynamic video segments. Prevailing fully-trained approaches are notoriously data-hungry. However, gathering large-scale STVG data is exceptionally challenging: dense frame-level bounding boxes and complex temporal language alignments are prohibitively expensive to annotate, especially for specialized video domains. Consequently, conventional models suffer from severe overfitting on these inherently limited datasets, while zero-shot foundational models lack the task-specific temporal awareness needed for precise localization. To resolve this small-data challenge, we introduce ST-GD, a data-efficient framework that adapts pre-trained 2D visual-language models (e.g., Grounding DINO) to video tasks. To avoid destroying pre-trained priors on small datasets, ST-GD keeps the base model frozen and strategically injects lightweight adapters (~10M trainable parameters) to instill spatio-temporal awareness, alongside a novel temporal decoder for boundary prediction. This design naturally counters data scarcity. Consequently, ST-GD excels in data-scarce scenarios, achieving highly competitive performance on the limited-scale HC-STVG v1/v2 benchmarks, while maintaining robust generalization on the VidSTG dataset. This validates ST-GD as a powerful paradigm for complex video understanding under strict small-data constraints.
Abstract (translated)
URL
https://arxiv.org/abs/2604.12346
PDF
-
All in One: A Unified Synthetic Data Pipeline for Multimodal Video Understanding
2026-04-14 06:17:35Tanzila Rahman, Renjie Liao, Leonid SigalAbstract
Training multimodal large language models (MLLMs) for video understanding requires large-scale annotated data spanning diverse tasks such as object counting, question answering, and segmentation. However, collecting and annotating multimodal video data in real-world is costly, slow, and inherently limited in diversity and coverage. To address this challenge, we propose a unified synthetic data generation pipeline capable of automatically producing unlimited multimodal video data with rich and diverse supervision. Our framework supports multiple task formats within a single pipeline, enabling scalable and consistent data creation across tasks. To further enhance reasoning ability, we introduce a VQA-based fine-tuning strategy that trains models to answer structured questions about visual content rather than relying solely on captions or simple instructions. This formulation encourages deeper visual grounding and reasoning. We evaluate our approach in three challenging tasks: video object counting, video-based visual question answering, and video object segmentation. Experimental results demonstrate that models trained predominantly on synthetic data generalize effectively to real-world datasets, often outperforming traditionally trained counterparts. Our findings highlight the potential of unified synthetic data pipelines as a scalable alternative to expensive real-world annotation for multimodal video understanding.
Abstract (translated)
URL
https://arxiv.org/abs/2604.12335
PDF
-
Style-Decoupled Adaptive Routing Network for Underwater Image Enhancement
2026-04-14 04:13:44Hang Xu, Chen Long, Bing Wang, Hao Chen, Zhen DongAbstract
Underwater Image Enhancement (UIE) is essential for robust visual perception in marine applications. However, existing methods predominantly rely on uniform mapping tailored to average dataset distributions, leading to over-processing mildly degraded images or insufficient recovery for severe ones. To address this challenge, we propose a novel adaptive enhancement framework, SDAR-Net. Unlike existing uniform paradigms, it first decouples specific degradation styles from the input and subsequently modulates the enhancement process adaptively. Specifically, since underwater degradation primarily shifts the appearance while keeping the scene structure, SDAR-Net formulates image features into dynamic degradation style embeddings and static scene structural representations through a carefully designed training framework. Subsequently, we introduce an adaptive routing mechanism. By evaluating style features and adaptively predicting soft weights at different enhancement states, it guides the weighted fusion of the corresponding image representations, accurately satisfying the adaptive restoration demands of each image. Extensive experiments show that SDAR-Net achieves a new state-of-the-art (SOTA) performance with a PSNR of 25.72 dB on real-world benchmark, and demonstrates its utility in downstream vision tasks. Our code is available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2604.12257
PDF
-
Nucleus-Image: Sparse MoE for Image Generation
2026-04-14 00:43:23Chandan Akiti, Ajay Modukuri, Murali Nandan Nagarapu, Gunavardhan Akiti, Haozhe LiuAbstract
We present Nucleus-Image, a text-to-image generation model that establishes a new Pareto frontier in quality-versus-efficiency by matching or exceeding leading models on GenEval, DPG-Bench, and OneIG-Bench while activating only approximately 2B parameters per forward pass. Nucleus-Image employs a sparse mixture-of-experts (MoE) diffusion transformer architecture with Expert-Choice Routing that scales total model capacity to 17B parameters across 64 routed experts per layer. We adopt a streamlined architecture optimized for inference efficiency by excluding text tokens from the transformer backbone entirely and using joint attention that enables text KV sharing across timesteps. To improve routing stability when using timestep modulation, we introduce a decoupled routing design that separates timestep-aware expert assignment from timestep-conditioned expert computation. We construct a large-scale training corpus of 1.5B high-quality training pairs spanning 700M unique images through multi-stage filtering, deduplication, aesthetic tiering, and caption curation. Training follows a progressive resolution curriculum (256 to 512 to 1024) with multi-aspect-ratio bucketing at every stage, coupled with progressive sparsification of the expert capacity factor. We adopt the Muon optimizer and share our parameter grouping recipe tailored for diffusion models with timestep modulation. Nucleus-Image demonstrates that sparse MoE scaling is a highly effective path to high-quality image generation, reaching the performance of models with significantly larger active parameter budgets at a fraction of the inference cost. These results are achieved without post-training optimization of any kind: no reinforcement learning, no direct preference optimization, and no human preference tuning. We release the training recipe, making Nucleus-Image the first fully open-source MoE diffusion model at this quality.
Abstract (translated)
URL
https://arxiv.org/abs/2604.12163
PDF
-
ViLL-E: Video LLM Embeddings for Retrieval
2026-04-13 23:54:58Rohit Gupta, Jayakrishnan Unnikrishnan, Fan Fei, Sheng Liu, Son Tran, Mubarak ShahAbstract
Video Large Language Models (VideoLLMs) excel at video understanding tasks where outputs are textual, such as Video Question Answering and Video Captioning. However, they underperform specialized embedding-based models in Retrieval tasks, such as Text-toVideo Retrieval and Moment Retrieval. We introduce ViLL-E (Video-LLM-Embed), a unified VideoLLM architecture endowed with a novel embedding generation mechanism that allows the model to "think longer" for complex videos and stop early for easy ones. We train this model with a three-stage training methodology combining generative and contrastive learning: initial large-scale pre-training with video-caption pairs; followed by continual training on a smaller, detailed-caption dataset; and concluding with task-specific fine-tuning on a novel multi-task dataset covering Video QA, Temporal Localization, Video Retrieval, and Video-Text Matching. Our model significantly improves temporal localization (on avg. 7% over other VideoLLMs) and video retrieval (up to 4% over dual encoder models), achieving performance comparable to state-of-the-art specialized embedding models while remaining competitive on VideoQA tasks. Furthermore, our joint contrastive-generative training unlocks new zero-shot capabilities, significantly outperforming state-of-the-art methods in composed video retrieval (+5% over SotA) and retrieval from long text (+2% over SotA).
Abstract (translated)
URL
https://arxiv.org/abs/2604.12148
PDF
-
TIPSv2: Advancing Vision-Language Pretraining with Enhanced Patch-Text Alignment
2026-04-13 20:00:04Bingyi Cao, Koert Chen, Kevis-Kokitsi Maninis, Kaifeng Chen, Arjun Karpur, Ye Xia, Sahil Dua, Tanmaya Dabral, Guangxing Han, Bohyung Han, Joshua Ainslie, Alex Bewley, Mithun Jacob, Ren\'e Wagner, Washington Ramos, Krzysztof Choromanski, Mojtaba Seyedhosseini, Howard Zhou, Andr\'e AraujoAbstract
Recent progress in vision-language pretraining has enabled significant improvements to many downstream computer vision applications, such as classification, retrieval, segmentation and depth prediction. However, a fundamental capability that these models still struggle with is aligning dense patch representations with text embeddings of corresponding concepts. In this work, we investigate this critical issue and propose novel techniques to enhance this capability in foundational vision-language models. First, we reveal that a patch-level distillation procedure significantly boosts dense patch-text alignment -- surprisingly, the patch-text alignment of the distilled student model strongly surpasses that of the teacher model. This observation inspires us to consider modifications to pretraining recipes, leading us to propose iBOT++, an upgrade to the commonly-used iBOT masked image objective, where unmasked tokens also contribute directly to the loss. This dramatically enhances patch-text alignment of pretrained models. Additionally, to improve vision-language pretraining efficiency and effectiveness, we modify the exponential moving average setup in the learning recipe, and introduce a caption sampling strategy to benefit from synthetic captions at different granularities. Combining these components, we develop TIPSv2, a new family of image-text encoder models suitable for a wide range of downstream applications. Through comprehensive experiments on 9 tasks and 20 datasets, we demonstrate strong performance, generally on par with or better than recent vision encoder models. Code and models are released via our project page at this https URL .
Abstract (translated)
URL
https://arxiv.org/abs/2604.12012
PDF
