Daily Papers

Short-form video poses new challenges to the quality assessment of user-generated content (UGC) due to its complex generation pipeline, rapid content variation, and mixed distortions. To address this challenge, we propose an end-to-end video quality assessment (VQA) framework that employs a dense visual encoder based on CLIP, and incorporates compression priors derived from the frequency domain to generate artifact- and structure-aware weight maps for feature aggregation. By explicitly decomposing artifact, structure, and original visual feature branches and adaptively fusing them over time through a learned gating module, the proposed method achieves accurate and efficient quality prediction. Experimental results show that our method achieves strong performance on short-form video datasets in terms of average rank and linear correlation (SRCC: 0.736, PLCC: 0.787), while maintaining efficient inference runtime. The code and additional results are available at: https://github.com/xinyiW915/FGSVQA.

Despite decades of progress, a truly input-size agnostic visual encoder-a fundamental characteristic of human vision-has remained elusive. We address this limitation by proposing MambaEye, a novel, causal sequential encoder that leverages the low complexity and causal-process based pure Mamba2 backbone. Unlike previous Mamba-based vision encoders that often employ bidirectional processing, our strictly unidirectional approach preserves the inherent causality of State Space Models, enabling the model to generate a prediction at any point in its input sequence. A core innovation is our use of relative move embedding, which encodes the spatial shift between consecutive patches, providing a strong inductive bias for translation invariance and making the model inherently adaptable to arbitrary image resolutions and scanning patterns. To achieve this, we introduce a novel diffusion-inspired loss function that provides dense, step-wise supervision, training the model to build confidence as it gathers more visual evidence. We demonstrate that MambaEye exhibits robust performance across a wide range of image resolutions, especially at higher resolutions such as 1536^2 on the ImageNet-1K classification task. This feat is achieved while maintaining linear time and memory complexity relative to the number of patches.

In recent years, notable advancements have been made in the domain of visual document understanding, with the prevailing architecture comprising a cascade of vision and language models. The text component can either be extracted explicitly with the use of external OCR models in OCR-based approaches, or alternatively, the vision model can be endowed with reading capabilities in OCR-free approaches. Typically, the queries to the model are input exclusively to the language component, necessitating the visual features to encompass the entire document. In this paper, we present VisFocus, an OCR-free method designed to better exploit the vision encoder's capacity by coupling it directly with the language prompt. To do so, we replace the down-sampling layers with layers that receive the input prompt and allow highlighting relevant parts of the document, while disregarding others. We pair the architecture enhancements with a novel pre-training task, using language masking on a snippet of the document text fed to the visual encoder in place of the prompt, to empower the model with focusing capabilities. Consequently, VisFocus learns to allocate its attention to text patches pertinent to the provided prompt. Our experiments demonstrate that this prompt-guided visual encoding approach significantly improves performance, achieving state-of-the-art results on various benchmarks.

Do we fully leverage the potential of visual encoder in Multimodal Large Language Models (MLLMs)? The recent outstanding performance of MLLMs in multimodal understanding has garnered broad attention from both academia and industry. In the current MLLM rat race, the focus seems to be predominantly on the linguistic side. We witness the rise of larger and higher-quality instruction datasets, as well as the involvement of larger-sized LLMs. Yet, scant attention has been directed towards the visual signals utilized by MLLMs, often assumed to be the final high-level features extracted by a frozen visual encoder. In this paper, we introduce the Dense Connector - a simple, effective, and plug-and-play vision-language connector that significantly enhances existing MLLMs by leveraging multi-layer visual features, with minimal additional computational overhead. Furthermore, our model, trained solely on images, showcases remarkable zero-shot capabilities in video understanding as well. Experimental results across various vision encoders, image resolutions, training dataset scales, varying sizes of LLMs (2.7B->70B), and diverse architectures of MLLMs (e.g., LLaVA and Mini-Gemini) validate the versatility and scalability of our approach, achieving state-of-the-art performance on across 19 image and video benchmarks. We hope that this work will provide valuable experience and serve as a basic module for future MLLM development.

Recent vision-language models (VLMs) typically rely on a single vision encoder trained with contrastive image-text objectives, such as CLIP-style pretraining. While contrastive encoders are effective for cross-modal alignment and retrieval, self-supervised visual encoders often capture richer dense semantics and exhibit stronger robustness on recognition and understanding tasks. In this work, we investigate how to scale the fusion of these complementary visual representations for vision-language modeling. We propose CoME-VL: Complementary Multi-Encoder Vision-Language, a modular fusion framework that integrates a contrastively trained vision encoder with a self-supervised DINO encoder. Our approach performs representation-level fusion by (i) entropy-guided multi-layer aggregation with orthogonality-constrained projections to reduce redundancy, and (ii) RoPE-enhanced cross-attention to align heterogeneous token grids and produce compact fused visual tokens. The fused tokens can be injected into a decoder-only LLM with minimal changes to standard VLM pipelines. Extensive experiments across diverse vision-language benchmarks demonstrate that CoME-VL consistently outperforms single-encoder baselines. In particular, we observe an average improvement of 4.9% on visual understanding tasks and 5.4% on grounding tasks. Our method achieves state-of-the-art performance on RefCOCO for detection while improving over the baseline by a large margin. Finally, we conduct ablation studies on layer merging, non-redundant feature mixing, and fusion capacity to evaluate how complementary contrastive and self-supervised signals affect VLM performance.

Despite recent advances, single-image super-resolution (SR) remains challenging, especially in real-world scenarios with complex degradations. Diffusion-based SR methods, particularly those built on Stable Diffusion, leverage strong generative priors but commonly rely on text conditioning derived from semantic captioning. Such textual descriptions provide only high-level semantics and lack the spatially aligned visual information required for faithful restoration, leading to a representation gap between abstract semantics and spatially aligned visual details. To address this limitation, we propose GramSR, a one-step diffusion-based SR framework that replaces text conditioning with dense visual features extracted from the low-resolution input using a pre-trained DINOv3 encoder. GramSR adopts a three-stage LoRA architecture, where pixel-level, semantic-level, and texture-level LoRA modules are trained sequentially. The pixel-level module focuses on degradation removal using ell_2 loss, the semantic-level module enhances perceptual details via LPIPS and CSD losses, and the texture-level module enforces feature correlation consistency through a Gram matrix loss computed from DINOv3 features. At inference, independent guidance scales enable flexible control over degradation removal, semantic enhancement, and texture preservation. Extensive experiments on standard SR benchmarks demonstrate that GramSR consistently outperforms existing one-step diffusion-based methods, achieving superior structural fidelity and texture realism. The code for this work is available at: https://github.com/aimagelab/GramSR.

Current foundation models for 3D shapes excel at global tasks (retrieval, classification) but transfer poorly to local part-level reasoning. Recent approaches leverage vision and language foundation models to directly solve dense tasks through multi-view renderings and text queries. While promising, these pipelines require expensive inference over multiple renderings, depend heavily on large language-model (LLM) prompt engineering for captions, and fail to exploit the inherent 3D geometry of shapes. We address this gap by introducing an encoder-only 3D model that produces language-aligned patch-level features directly from point clouds. Our pre-training approach builds on existing data engines that generate part-annotated 3D shapes by pairing multi-view SAM regions with VLM captioning. Using this data, we train a point cloud transformer encoder in two stages: (1) distillation of dense 2D features from visual encoders such as DINOv2 into 3D patches, and (2) alignment of these patch embeddings with part-level text embeddings through a multi-positive contrastive objective. Our 3D encoder achieves zero-shot 3D part segmentation with fast single-pass inference without any test-time multi-view rendering, while significantly outperforming previous rendering-based and feed-forward approaches across several 3D part segmentation benchmarks. Project website: https://souhail-hadgi.github.io/patchalign3dsite/

Visual representations are central to the learning and generalization capabilities of robotic manipulation policies. While existing methods rely on global or dense features, such representations often entangle task-relevant and irrelevant scene information, limiting robustness under distribution shifts. In this work, we investigate object-centric representations (OCR) as a structured alternative that segments visual input into a finished set of entities, introducing inductive biases that align more naturally with manipulation tasks. We benchmark a range of visual encoders-object-centric, global and dense methods-across a suite of simulated and real-world manipulation tasks ranging from simple to complex, and evaluate their generalization under diverse visual conditions including changes in lighting, texture, and the presence of distractors. Our findings reveal that OCR-based policies outperform dense and global representations in generalization settings, even without task-specific pretraining. These insights suggest that OCR is a promising direction for designing visual systems that generalize effectively in dynamic, real-world robotic environments.

Monocular depth estimation aims to recover the depth information of 3D scenes from 2D images. Recent work has made significant progress, but its reliance on large-scale datasets and complex decoders has limited its efficiency and generalization ability. In this paper, we propose a lightweight and data-centric framework for zero-shot monocular depth estimation. We first adopt DINOv3 as the visual encoder to obtain high-quality dense features. Secondly, to address the inherent drawbacks of the complex structure of the DPT, we design the Simple Depth Transformer (SDT), a compact transformer-based decoder. Compared to the DPT, it uses a single-path feature fusion and upsampling process to reduce the computational overhead of cross-scale feature fusion, achieving higher accuracy while reducing the number of parameters by approximately 85%-89%. Furthermore, we propose a quality-based filtering strategy to filter out harmful samples, thereby reducing dataset size while improving overall training quality. Extensive experiments on five benchmarks demonstrate that our framework surpasses the DPT in accuracy. This work highlights the importance of balancing model design and data quality for achieving efficient and generalizable zero-shot depth estimation. Code: https://github.com/AIGeeksGroup/AnyDepth. Website: https://aigeeksgroup.github.io/AnyDepth.

Recent progress in vision Transformers exhibits great success in various tasks driven by the new spatial modeling mechanism based on dot-product self-attention. In this paper, we show that the key ingredients behind the vision Transformers, namely input-adaptive, long-range and high-order spatial interactions, can also be efficiently implemented with a convolution-based framework. We present the Recursive Gated Convolution (g^nConv) that performs high-order spatial interactions with gated convolutions and recursive designs. The new operation is highly flexible and customizable, which is compatible with various variants of convolution and extends the two-order interactions in self-attention to arbitrary orders without introducing significant extra computation. g^nConv can serve as a plug-and-play module to improve various vision Transformers and convolution-based models. Based on the operation, we construct a new family of generic vision backbones named HorNet. Extensive experiments on ImageNet classification, COCO object detection and ADE20K semantic segmentation show HorNet outperform Swin Transformers and ConvNeXt by a significant margin with similar overall architecture and training configurations. HorNet also shows favorable scalability to more training data and larger model sizes. Apart from the effectiveness in visual encoders, we also show g^nConv can be applied to task-specific decoders and consistently improve dense prediction performance with less computation. Our results demonstrate that g^nConv can be a new basic module for visual modeling that effectively combines the merits of both vision Transformers and CNNs. Code is available at https://github.com/raoyongming/HorNet

In this paper, we tackle an emerging computer vision task, open-vocabulary universal image segmentation, that aims to perform semantic/instance/panoptic segmentation (background semantic labeling + foreground instance segmentation) for arbitrary categories of text-based descriptions in inference time. We first build a baseline method by directly adopting pre-trained CLIP models without finetuning or distillation. We then develop MaskCLIP, a Transformer-based approach with a MaskCLIP Visual Encoder, which is an encoder-only module that seamlessly integrates mask tokens with a pre-trained ViT CLIP model for semantic/instance segmentation and class prediction. MaskCLIP learns to efficiently and effectively utilize pre-trained partial/dense CLIP features within the MaskCLIP Visual Encoder that avoids the time-consuming student-teacher training process. MaskCLIP outperforms previous methods for semantic/instance/panoptic segmentation on ADE20K and PASCAL datasets. We show qualitative illustrations for MaskCLIP with online custom categories. Project website: https://maskclip.github.io.

Motion capture now underpins content creation far beyond digital humans, yet most existing pipelines remain species- or template-specific. We formalize this gap as Category-Agnostic Motion Capture (CAMoCap): given a monocular video and an arbitrary rigged 3D asset as a prompt, the goal is to reconstruct a rotation-based animation such as BVH that directly drives the specific asset. We present MoCapAnything, a reference-guided, factorized framework that first predicts 3D joint trajectories and then recovers asset-specific rotations via constraint-aware inverse kinematics. The system contains three learnable modules and a lightweight IK stage: (1) a Reference Prompt Encoder that extracts per-joint queries from the asset's skeleton, mesh, and rendered images; (2) a Video Feature Extractor that computes dense visual descriptors and reconstructs a coarse 4D deforming mesh to bridge the gap between video and joint space; and (3) a Unified Motion Decoder that fuses these cues to produce temporally coherent trajectories. We also curate Truebones Zoo with 1038 motion clips, each providing a standardized skeleton-mesh-render triad. Experiments on both in-domain benchmarks and in-the-wild videos show that MoCapAnything delivers high-quality skeletal animations and exhibits meaningful cross-species retargeting across heterogeneous rigs, enabling scalable, prompt-driven 3D motion capture for arbitrary assets. Project page: https://animotionlab.github.io/MoCapAnything/

Zero-shot Semantic Segmentation (ZSS) aims to segment both seen and unseen classes using supervision from only seen classes. Beyond adaptation-based methods, distillation-based approaches transfer vision-language alignment of vision-language model, e.g., CLIP, to segmentation models. However, such knowledge transfer remains challenging due to: (1) the difficulty of aligning vision-based features with the textual space, which requires combining spatial precision with vision-language alignment; and (2) the semantic gap between CLIP's global representations and the local, fine-grained features of segmentation models. To address challenge (1), we propose Chimera-Seg, which integrates a segmentation backbone as the body and a CLIP-based semantic head as the head, like the Chimera in Greek mythology, combining spatial precision with vision-language alignment. Specifically, Chimera-Seg comprises a trainable segmentation model and a CLIP Semantic Head (CSH), which maps dense features into the CLIP-aligned space. The CSH incorporates a frozen subnetwork and fixed projection layers from the CLIP visual encoder, along with lightweight trainable components. The partial module from CLIP visual encoder, paired with the segmentation model, retains segmentation capability while easing the mapping to CLIP's semantic space. To address challenge (2), we propose Selective Global Distillation (SGD), which distills knowledge from dense features exhibiting high similarity to the CLIP CLS token, while gradually reducing the number of features used for alignment as training progresses. Besides, we also use a Semantic Alignment Module (SAM) to further align dense visual features with semantic embeddings extracted from the frozen CLIP text encoder. Experiments on two benchmarks show improvements of 0.9% and 1.2% in hIoU.

Visuomotor robot policies, increasingly pre-trained on large-scale datasets, promise significant advancements across robotics domains. However, aligning these policies with end-user preferences remains a challenge, particularly when the preferences are hard to specify. While reinforcement learning from human feedback (RLHF) has become the predominant mechanism for alignment in non-embodied domains like large language models, it has not seen the same success in aligning visuomotor policies due to the prohibitive amount of human feedback required to learn visual reward functions. To address this limitation, we propose Representation-Aligned Preference-based Learning (RAPL), an observation-only method for learning visual rewards from significantly less human preference feedback. Unlike traditional RLHF, RAPL focuses human feedback on fine-tuning pre-trained vision encoders to align with the end-user's visual representation and then constructs a dense visual reward via feature matching in this aligned representation space. We first validate RAPL through simulation experiments in the X-Magical benchmark and Franka Panda robotic manipulation, demonstrating that it can learn rewards aligned with human preferences, more efficiently uses preference data, and generalizes across robot embodiments. Finally, our hardware experiments align pre-trained Diffusion Policies for three object manipulation tasks. We find that RAPL can fine-tune these policies with 5x less real human preference data, taking the first step towards minimizing human feedback while maximizing visuomotor robot policy alignment.

Unified multimodal models (UMMs) unify visual understanding and generation within a single architecture. However, conventional training relies on image-text pairs (or sequences) whose captions are typically sparse and miss fine-grained visual details--even when they use hundreds of words to describe a simple image. We introduce Reconstruction Alignment (RecA), a resource-efficient post-training method that leverages visual understanding encoder embeddings as dense "text prompts," providing rich supervision without captions. Concretely, RecA conditions a UMM on its own visual understanding embeddings and optimizes it to reconstruct the input image with a self-supervised reconstruction loss, thereby realigning understanding and generation. Despite its simplicity, RecA is broadly applicable: across autoregressive, masked-autoregressive, and diffusion-based UMMs, it consistently improves generation and editing fidelity. With only 27 GPU-hours, post-training with RecA substantially improves image generation performance on GenEval (0.73rightarrow0.90) and DPGBench (80.93rightarrow88.15), while also boosting editing benchmarks (ImgEdit 3.38rightarrow3.75, GEdit 6.94rightarrow7.25). Notably, RecA surpasses much larger open-source models and applies broadly across diverse UMM architectures, establishing it as an efficient and general post-training alignment strategy for UMMs

Recent advances in omni-modal large language models have enabled remarkable progress in joint vision-audio understanding. However, prevailing architectures rely on modality-specific encoders with a video-coarse, audio-dense design -- sampling visual frames at 1--2 fps while processing audio waveforms at 25 fps -- resulting in systems that perceive video frame by frame, modality by modality rather than holistically as humans do. Such a discrepancy leaves models with impoverished cross-modal interaction during encoding and an inability to capture fine-grained visual motion. To bridge this gap, we present Omni-Encoder, a unified Transformer backbone designed to co-embed visual and audio signals at a symmetrical 25 fps within a shared latent space. This architecture leverages three core innovations -- the Omni-Encoder Token Template, Omni-RoPE, and Temporal Window Shifting -- to effectively reconcile the dual challenges of modality disentanglement and computational efficiency. Experiments demonstrate that, compared to the modality-specific baseline Qwen2.5-Omni under the same input token budget to the LLM decoder, Omni-Encoder delivers substantial gains on visual continuous understanding tasks -- such as sign language recognition and fine-grained sports action analysis -- while maintaining competitive performance on established audio-visual benchmarks such as AVQA and Speaker Identification and Localization. These results suggest that unified omnivorous encoding offers a promising direction for building omni-modal models that more closely reflect the integrated nature of human perception.

We introduce DIP, a novel unsupervised post-training method designed to enhance dense image representations in large-scale pretrained vision encoders for in-context scene understanding. Unlike prior approaches that rely on complex self-distillation architectures, our method trains the vision encoder using pseudo-tasks that explicitly simulate downstream in-context scenarios, inspired by meta-learning principles. To enable post-training on unlabeled data, we propose an automatic mechanism for generating in-context tasks that combines a pretrained diffusion model and the vision encoder itself. DIP is simple, unsupervised, and computationally efficient, requiring less than 9 hours on a single A100 GPU. By learning dense representations through pseudo in-context tasks, it achieves strong performance across a wide variety of downstream real-world in-context scene understanding tasks. It outperforms both the initial vision encoder and prior methods, offering a practical and effective solution for improving dense representations. Code available here: https://github.com/sirkosophia/DIP

We introduce Perception Encoder (PE), a state-of-the-art encoder for image and video understanding trained via simple vision-language learning. Traditionally, vision encoders have relied on a variety of pretraining objectives, each tailored to specific downstream tasks such as classification, captioning, or localization. Surprisingly, after scaling our carefully tuned image pretraining recipe and refining with our robust video data engine, we find that contrastive vision-language training alone can produce strong, general embeddings for all of these downstream tasks. There is only one caveat: these embeddings are hidden within the intermediate layers of the network. To draw them out, we introduce two alignment methods, language alignment for multimodal language modeling, and spatial alignment for dense prediction. Together with the core contrastive checkpoint, our PE family of models achieves state-of-the-art performance on a wide variety of tasks, including zero-shot image and video classification and retrieval; document, image, and video Q&A; and spatial tasks such as detection, depth estimation, and tracking. To foster further research, we are releasing our models, code, and a novel dataset of synthetically and human-annotated videos.

Knowledge-Intensive Visual Question Answering (KI-VQA) refers to answering a question about an image whose answer does not lie in the image. This paper presents a new pipeline for KI-VQA tasks, consisting of a retriever and a reader. First, we introduce DEDR, a symmetric dual encoding dense retrieval framework in which documents and queries are encoded into a shared embedding space using uni-modal (textual) and multi-modal encoders. We introduce an iterative knowledge distillation approach that bridges the gap between the representation spaces in these two encoders. Extensive evaluation on two well-established KI-VQA datasets, i.e., OK-VQA and FVQA, suggests that DEDR outperforms state-of-the-art baselines by 11.6% and 30.9% on OK-VQA and FVQA, respectively. Utilizing the passages retrieved by DEDR, we further introduce MM-FiD, an encoder-decoder multi-modal fusion-in-decoder model, for generating a textual answer for KI-VQA tasks. MM-FiD encodes the question, the image, and each retrieved passage separately and uses all passages jointly in its decoder. Compared to competitive baselines in the literature, this approach leads to 5.5% and 8.5% improvements in terms of question answering accuracy on OK-VQA and FVQA, respectively.

We introduce SigLIP 2, a family of new multilingual vision-language encoders that build on the success of the original SigLIP. In this second iteration, we extend the original image-text training objective with several prior, independently developed techniques into a unified recipe -- this includes captioning-based pretraining, self-supervised losses (self-distillation, masked prediction) and online data curation. With these changes, SigLIP 2 models outperform their SigLIP counterparts at all model scales in core capabilities, including zero-shot classification, image-text retrieval, and transfer performance when extracting visual representations for Vision-Language Models (VLMs). Furthermore, the new training recipe leads to significant improvements on localization and dense prediction tasks. We also train variants which support multiple resolutions and preserve the input's native aspect ratio. Finally, we train on a more diverse data-mixture that includes de-biasing techniques, leading to much better multilingual understanding and improved fairness. To allow users to trade off inference cost with performance, we release model checkpoints at four sizes: ViT-B (86M), L (303M), So400m (400M), and g (1B).

Pre-trained encoders are widely employed in dense prediction tasks for their capability to effectively extract visual features from images. The decoder subsequently processes these features to generate pixel-level predictions. However, due to structural differences and variations in input data, only encoders benefit from pre-learned representations from vision benchmarks such as image classification and self-supervised learning, while decoders are typically trained from scratch. In this paper, we introduce timesNet, which facilitates a "pre-trained encoder times pre-trained decoder" collaboration through three innovative designs. timesNet enables the direct utilization of pre-trained models within the decoder, integrating pre-learned representations into the decoding process to enhance performance in dense prediction tasks. By simply coupling the pre-trained encoder and pre-trained decoder, timesNet distinguishes itself as a highly promising approach. Remarkably, it achieves this without relying on decoding-specific structures or task-specific algorithms. Despite its streamlined design, timesNet outperforms advanced methods in tasks such as monocular depth estimation and semantic segmentation, achieving state-of-the-art performance particularly in monocular depth estimation. and semantic segmentation, achieving state-of-the-art results, especially in monocular depth estimation. embedding algorithms. Despite its streamlined design, timesNet outperforms advanced methods in tasks such as monocular depth estimation and semantic segmentation, achieving state-of-the-art performance particularly in monocular depth estimation.

Despite the remarkable success of large-scale pre-trained image representation models (i.e., vision encoders) across various vision tasks, they are predominantly trained on 2D image data and therefore often fail to capture 3D spatial relationships between objects and backgrounds in the real world, constraining their effectiveness in many downstream applications. To address this, we propose SpatialBoost, a scalable framework that enhances the spatial awareness of existing pre-trained vision encoders by injecting 3D spatial knowledge expressed in linguistic descriptions. The core idea involves converting dense 3D spatial information from 2D images into linguistic expressions, which is then used to inject such spatial knowledge into vision encoders through a Large Language Model (LLM). To this end, we adopt a multi-turn Chain-of-Thought (CoT) reasoning process that progressively incorporates dense spatial knowledge and builds hierarchical spatial understanding. To validate effectiveness, we adapt SpatialBoost to state-of-the-art vision encoders such as DINOv3, and evaluate its performance gains on a wide range of benchmarks requiring both 3D perception and general vision abilities. For instance, SpatialBoost improves DINOv3 performance from 55.9 to 59.7 mIoU on ADE20K, achieving state-of-the-art performance with 3.8% gain over the pre-trained DINOv3.

Composed video retrieval is a challenging task that strives to retrieve a target video based on a query video and a textual description detailing specific modifications. Standard retrieval frameworks typically struggle to handle the complexity of fine-grained compositional queries and variations in temporal understanding limiting their retrieval ability in the fine-grained setting. To address this issue, we introduce a novel dataset that captures both fine-grained and composed actions across diverse video segments, enabling more detailed compositional changes in retrieved video content. The proposed dataset, named Dense-WebVid-CoVR, consists of 1.6 million samples with dense modification text that is around seven times more than its existing counterpart. We further develop a new model that integrates visual and textual information through Cross-Attention (CA) fusion using grounded text encoder, enabling precise alignment between dense query modifications and target videos. The proposed model achieves state-of-the-art results surpassing existing methods on all metrics. Notably, it achieves 71.3\% Recall@1 in visual+text setting and outperforms the state-of-the-art by 3.4\%, highlighting its efficacy in terms of leveraging detailed video descriptions and dense modification texts. Our proposed dataset, code, and model are available at :https://github.com/OmkarThawakar/BSE-CoVR

Despite recent progress, vision-language encoders struggle with two core limitations: (1) weak alignment between language and dense vision features, which hurts tasks like open-vocabulary semantic segmentation; and (2) high token counts for fine-grained visual representations, which limits scalability to long videos. This work addresses both limitations. We propose T-REN (Text-aligned Region Encoder Network), an efficient encoder that maps visual data to a compact set of text-aligned region-level representations (or region tokens). T-REN achieves this through a lightweight network added on top of a frozen vision backbone, trained to pool patch-level representations within each semantic region into region tokens and align them with region-level text annotations. With only 3.7% additional parameters compared to the vision-language backbone, this design yields substantially stronger dense cross-modal understanding while reducing the token count by orders of magnitude. Specifically, T-REN delivers +5.9 mIoU on ADE20K open-vocabulary segmentation, +18.4% recall on COCO object-level text-image retrieval, +15.6% recall on Ego4D video object localization, and +17.6% mIoU on VSPW video scene parsing, all while reducing token counts by more than 24x for images and 187x for videos compared to the patch-based vision-language backbone. The code and model are available at https://github.com/savya08/T-REN.

Vision Language Model (VLM) development has largely relied on scaling model size, which hinders deployment on compute-constrained mobile and edge devices such as smartphones and robots. In this work, we explore the performance limits of compact (e.g., 2B and 8B) VLMs. We challenge the prevailing practice that state-of-the-art VLMs must rely on vision encoders initialized via massive contrastive pretraining (e.g., CLIP/SigLIP). We identify an objective mismatch: contrastive learning, optimized for discrimination, enforces coarse and category-level invariances that suppress fine-grained visual cues needed for dense captioning and complex VLM reasoning. To address this issue, we present Penguin-VL, whose vision encoder is initialized from a text-only LLM. Our experiments reveal that Penguin-Encoder serves as a superior alternative to traditional contrastive pretraining, unlocking a higher degree of visual fidelity and data efficiency for multimodal understanding. Across various image and video benchmarks, Penguin-VL achieves performance comparable to leading VLMs (e.g., Qwen3-VL) in mathematical reasoning and surpasses them in tasks such as document understanding, visual knowledge, and multi-perspective video understanding. Notably, these gains are achieved with a lightweight architecture, demonstrating that improved visual representation rather than model scaling is the primary driver of performance. Our ablations show that Penguin-Encoder consistently outperforms contrastive-pretrained encoders, preserving fine-grained spatial and temporal cues that are critical for dense perception and complex reasoning. This makes it a strong drop-in alternative for compute-efficient VLMs and enables high performance in resource-constrained settings. Code: https://github.com/tencent-ailab/Penguin-VL

In the domain of computer vision, Parameter-Efficient Tuning (PET) is increasingly replacing the traditional paradigm of pre-training followed by full fine-tuning. PET is particularly favored for its effectiveness in large foundation models, as it streamlines transfer learning costs and optimizes hardware utilization. However, the current PET methods are mainly designed for single-modal optimization. While some pioneering studies have undertaken preliminary explorations, they still remain at the level of aligned encoders (e.g., CLIP) and lack exploration of misaligned encoders. These methods show sub-optimal performance with misaligned encoders, as they fail to effectively align the multimodal features during fine-tuning. In this paper, we introduce DETRIS, a parameter-efficient tuning framework designed to enhance low-rank visual feature propagation by establishing dense interconnections between each layer and all preceding layers, which enables effective cross-modal feature interaction and adaptation to misaligned encoders. We also suggest using text adapters to improve textual features. Our simple yet efficient approach greatly surpasses state-of-the-art methods with 0.9% to 1.8% backbone parameter updates, evaluated on challenging benchmarks. Our project is available at https://github.com/jiaqihuang01/DETRIS.

The core of video understanding tasks, such as recognition, captioning, and tracking, is to automatically detect objects or actions in a video and analyze their temporal evolution. Despite sharing a common goal, different tasks often rely on distinct model architectures and annotation formats. In contrast, natural language processing benefits from a unified output space, i.e., text sequences, which simplifies the training of powerful foundational language models, such as GPT-3, with extensive training corpora. Inspired by this, we seek to unify the output space of video understanding tasks by using languages as labels and additionally introducing time and box tokens. In this way, a variety of video tasks could be formulated as video-grounded token generation. This enables us to address various types of video tasks, including classification (such as action recognition), captioning (covering clip captioning, video question answering, and dense video captioning), and localization tasks (such as visual object tracking) within a fully shared encoder-decoder architecture, following a generative framework. Through comprehensive experiments, we demonstrate such a simple and straightforward idea is quite effective and can achieve state-of-the-art or competitive results on seven video benchmarks, providing a novel perspective for more universal video understanding. Code is available at https://github.com/wangjk666/OmniVid.

We introduce dense vision transformers, an architecture that leverages vision transformers in place of convolutional networks as a backbone for dense prediction tasks. We assemble tokens from various stages of the vision transformer into image-like representations at various resolutions and progressively combine them into full-resolution predictions using a convolutional decoder. The transformer backbone processes representations at a constant and relatively high resolution and has a global receptive field at every stage. These properties allow the dense vision transformer to provide finer-grained and more globally coherent predictions when compared to fully-convolutional networks. Our experiments show that this architecture yields substantial improvements on dense prediction tasks, especially when a large amount of training data is available. For monocular depth estimation, we observe an improvement of up to 28% in relative performance when compared to a state-of-the-art fully-convolutional network. When applied to semantic segmentation, dense vision transformers set a new state of the art on ADE20K with 49.02% mIoU. We further show that the architecture can be fine-tuned on smaller datasets such as NYUv2, KITTI, and Pascal Context where it also sets the new state of the art. Our models are available at https://github.com/intel-isl/DPT.

Sparsely-gated Mixture of Experts networks (MoEs) have demonstrated excellent scalability in Natural Language Processing. In Computer Vision, however, almost all performant networks are "dense", that is, every input is processed by every parameter. We present a Vision MoE (V-MoE), a sparse version of the Vision Transformer, that is scalable and competitive with the largest dense networks. When applied to image recognition, V-MoE matches the performance of state-of-the-art networks, while requiring as little as half of the compute at inference time. Further, we propose an extension to the routing algorithm that can prioritize subsets of each input across the entire batch, leading to adaptive per-image compute. This allows V-MoE to trade-off performance and compute smoothly at test-time. Finally, we demonstrate the potential of V-MoE to scale vision models, and train a 15B parameter model that attains 90.35% on ImageNet.

The key idea of current deep learning methods for dense prediction is to apply a model on a regular patch centered on each pixel to make pixel-wise predictions. These methods are limited in the sense that the patches are determined by network architecture instead of learned from data. In this work, we propose the dense transformer networks, which can learn the shapes and sizes of patches from data. The dense transformer networks employ an encoder-decoder architecture, and a pair of dense transformer modules are inserted into each of the encoder and decoder paths. The novelty of this work is that we provide technical solutions for learning the shapes and sizes of patches from data and efficiently restoring the spatial correspondence required for dense prediction. The proposed dense transformer modules are differentiable, thus the entire network can be trained. We apply the proposed networks on natural and biological image segmentation tasks and show superior performance is achieved in comparison to baseline methods.

Human visual recognition is a sparse process, where only a few salient visual cues are attended to rather than traversing every detail uniformly. However, most current vision networks follow a dense paradigm, processing every single visual unit (e.g,, pixel or patch) in a uniform manner. In this paper, we challenge this dense paradigm and present a new method, coined SparseFormer, to imitate human's sparse visual recognition in an end-to-end manner. SparseFormer learns to represent images using a highly limited number of tokens (down to 49) in the latent space with sparse feature sampling procedure instead of processing dense units in the original pixel space. Therefore, SparseFormer circumvents most of dense operations on the image space and has much lower computational costs. Experiments on the ImageNet classification benchmark dataset show that SparseFormer achieves performance on par with canonical or well-established models while offering better accuracy-throughput tradeoff. Moreover, the design of our network can be easily extended to the video classification with promising performance at lower computational costs. We hope that our work can provide an alternative way for visual modeling and inspire further research on sparse neural architectures. The code will be publicly available at https://github.com/showlab/sparseformer

We present DenseAV, a novel dual encoder grounding architecture that learns high-resolution, semantically meaningful, and audio-visually aligned features solely through watching videos. We show that DenseAV can discover the ``meaning'' of words and the ``location'' of sounds without explicit localization supervision. Furthermore, it automatically discovers and distinguishes between these two types of associations without supervision. We show that DenseAV's localization abilities arise from a new multi-head feature aggregation operator that directly compares dense image and audio representations for contrastive learning. In contrast, many other systems that learn ``global'' audio and video representations cannot localize words and sound. Finally, we contribute two new datasets to improve the evaluation of AV representations through speech and sound prompted semantic segmentation. On these and other datasets we show DenseAV dramatically outperforms the prior art on speech and sound prompted semantic segmentation. DenseAV outperforms the previous state-of-the-art, ImageBind, on cross-modal retrieval using fewer than half of the parameters. Project Page: https://aka.ms/denseav{https://aka.ms/denseav}

Recent dense audio-visual (AV) models achieve impressive retrieval and emergent localization, but almost all evidence comes from English-centric, caption-rich web video. It is unclear whether these objectives survive in low-resource, code-switched, and noisy multilingual settings that typify developing regions. We show they do**-**and that the choice of aggregation function becomes even more critical. Using a multilingual subset of Project Vaani spanning dozens of Indian languages and dialectal variants, we compare three contrastive objectives: (i) a global mean-pooled loss (CLIP-style), (ii) a dense max-mean token matcher (DenseAV-style), and (iii) a simple hybrid (motivated by frozen-vision alignment strategies). The dense objective delivers a +59% relative R@1 (Audio Visual) improvement over global pooling and substantially lower mean/median ranks, while consistently producing sharp zero-shot localization heatmaps of spoken objects-despite keeping the vision backbone entirely frozen (no LoRA / partial fine-tuning). Our results demonstrate that dense token routing is not a luxury of high-resource English corpora; it is more decisive when annotations and acoustic cleanliness are scarce. We release the codebase and trained models.

Dense and versatile image representations underpin the success of virtually all computer vision applications. However, state-of-the-art networks, such as transformers, produce low-resolution feature grids, which are suboptimal for dense prediction tasks. To address this limitation, we present FlowFeat, a high-resolution and multi-task feature representation. The key ingredient behind FlowFeat is a novel distillation technique that embeds a distribution of plausible apparent motions, or motion profiles. By leveraging optical flow networks and diverse video data, we develop an effective self-supervised training framework that statistically approximates the apparent motion. With its remarkable level of spatial detail, FlowFeat encodes a compelling degree of geometric and semantic cues while exhibiting high temporal consistency. Empirically, FlowFeat significantly enhances the representational power of five state-of-the-art encoders and alternative upsampling strategies across three dense tasks: video object segmentation, monocular depth estimation and semantic segmentation. Training FlowFeat is computationally inexpensive and robust to inaccurate flow estimation, remaining highly effective even when using unsupervised flow networks. Our work takes a step forward towards reliable and versatile dense image representations.

Video Language Models (VideoLMs) empower AI systems to understand temporal dynamics in videos. To fit to the maximum context window constraint, current methods use keyframe sampling which can miss both macro-level events and micro-level details due to the sparse temporal coverage. Furthermore, processing full images and their tokens for each frame incurs substantial computational overhead. To address these limitations, we propose to leverage video codec primitives (specifically motion vectors and residuals) which natively encode video redundancy and sparsity without requiring expensive full-image encoding for most frames. To this end, we introduce lightweight transformer-based encoders that aggregate codec primitives and align their representations with image encoder embeddings through a pre-training strategy that accelerates convergence during end-to-end fine-tuning. Our approach reduces the time-to-first-token by up to 86% and token usage by up to 93% compared to standard VideoLMs. Moreover, by varying the keyframe and codec primitive densities we are able to maintain or exceed performance on 14 diverse video understanding benchmarks spanning general question answering, temporal reasoning, long-form understanding, and spatial scene understanding.

Most existing vision encoders map images into a fixed-length sequence of tokens, overlooking the fact that different images contain varying amounts of information. For example, a visually complex image (e.g., a cluttered room) inherently carries more information and thus deserves more tokens than a simple image (e.g., a blank wall). To address this inefficiency, we propose DOVE, a dynamic vision encoder that produces a variable number of visual tokens (i.e., continuous representation vectors) to reconstruct each image. Our results show that DOVE significantly reduces the average number of tokens while maintaining high reconstruction quality. In several linear probing and downstream multimodal tasks, it outperforms existing autoencoder-based tokenization methods when using far fewer tokens, capturing more expressive semantic features compared to fixed-length encoding. We further extend DOVE with query-conditioned tokenization. By guiding the model to focus on query-relevant regions, it achieves more efficient and targeted semantic extraction. Our code and checkpoints are available at https://dove-encoder.github.io/dove-encoder.

Very recently, a variety of vision transformer architectures for dense prediction tasks have been proposed and they show that the design of spatial attention is critical to their success in these tasks. In this work, we revisit the design of the spatial attention and demonstrate that a carefully-devised yet simple spatial attention mechanism performs favourably against the state-of-the-art schemes. As a result, we propose two vision transformer architectures, namely, Twins-PCPVT and Twins-SVT. Our proposed architectures are highly-efficient and easy to implement, only involving matrix multiplications that are highly optimized in modern deep learning frameworks. More importantly, the proposed architectures achieve excellent performance on a wide range of visual tasks, including image level classification as well as dense detection and segmentation. The simplicity and strong performance suggest that our proposed architectures may serve as stronger backbones for many vision tasks. Our code is released at https://github.com/Meituan-AutoML/Twins .

Vision-language models (VLMs) are typically composed of a vision encoder, e.g. CLIP, and a language model (LM) that interprets the encoded features to solve downstream tasks. Despite remarkable progress, VLMs are subject to several shortcomings due to the limited capabilities of vision encoders, e.g. "blindness" to certain image features, visual hallucination, etc. To address these issues, we study broadening the visual encoding capabilities of VLMs. We first comprehensively benchmark several vision encoders with different inductive biases for solving VLM tasks. We observe that there is no single encoding configuration that consistently achieves top performance across different tasks, and encoders with different biases can perform surprisingly similarly. Motivated by this, we introduce a method, named BRAVE, that consolidates features from multiple frozen encoders into a more versatile representation that can be directly fed as the input to a frozen LM. BRAVE achieves state-of-the-art performance on a broad range of captioning and VQA benchmarks and significantly reduces the aforementioned issues of VLMs, while requiring a smaller number of trainable parameters than existing methods and having a more compressed representation. Our results highlight the potential of incorporating different visual biases for a more broad and contextualized visual understanding of VLMs.

Recent work has shown that convolutional networks can be substantially deeper, more accurate, and efficient to train if they contain shorter connections between layers close to the input and those close to the output. In this paper, we embrace this observation and introduce the Dense Convolutional Network (DenseNet), which connects each layer to every other layer in a feed-forward fashion. Whereas traditional convolutional networks with L layers have L connections - one between each layer and its subsequent layer - our network has L(L+1)/2 direct connections. For each layer, the feature-maps of all preceding layers are used as inputs, and its own feature-maps are used as inputs into all subsequent layers. DenseNets have several compelling advantages: they alleviate the vanishing-gradient problem, strengthen feature propagation, encourage feature reuse, and substantially reduce the number of parameters. We evaluate our proposed architecture on four highly competitive object recognition benchmark tasks (CIFAR-10, CIFAR-100, SVHN, and ImageNet). DenseNets obtain significant improvements over the state-of-the-art on most of them, whilst requiring less computation to achieve high performance. Code and pre-trained models are available at https://github.com/liuzhuang13/DenseNet .

In this work, we introduce a deep-learning framework designed for estimating dense image correspondences. Our fully convolutional model generates dense feature maps for images, where each pixel is associated with a descriptor that can be matched across multiple images. Unlike previous methods, our model is trained on synthetic data that includes significant distortions, such as perspective changes, illumination variations, shadows, and specular highlights. Utilizing contrastive learning, our feature maps achieve greater invariance to these distortions, enabling robust matching. Notably, our method eliminates the need for a keypoint detector, setting it apart from many existing image-matching techniques.

This paper revives Densely Connected Convolutional Networks (DenseNets) and reveals the underrated effectiveness over predominant ResNet-style architectures. We believe DenseNets' potential was overlooked due to untouched training methods and traditional design elements not fully revealing their capabilities. Our pilot study shows dense connections through concatenation are strong, demonstrating that DenseNets can be revitalized to compete with modern architectures. We methodically refine suboptimal components - architectural adjustments, block redesign, and improved training recipes towards widening DenseNets and boosting memory efficiency while keeping concatenation shortcuts. Our models, employing simple architectural elements, ultimately surpass Swin Transformer, ConvNeXt, and DeiT-III - key architectures in the residual learning lineage. Furthermore, our models exhibit near state-of-the-art performance on ImageNet-1K, competing with the very recent models and downstream tasks, ADE20k semantic segmentation, and COCO object detection/instance segmentation. Finally, we provide empirical analyses that uncover the merits of the concatenation over additive shortcuts, steering a renewed preference towards DenseNet-style designs. Our code is available at https://github.com/naver-ai/rdnet.

Estimating accurate depth from a single image is challenging because it is an ill-posed problem as infinitely many 3D scenes can be projected to the same 2D scene. However, recent works based on deep convolutional neural networks show great progress with plausible results. The convolutional neural networks are generally composed of two parts: an encoder for dense feature extraction and a decoder for predicting the desired depth. In the encoder-decoder schemes, repeated strided convolution and spatial pooling layers lower the spatial resolution of transitional outputs, and several techniques such as skip connections or multi-layer deconvolutional networks are adopted to recover the original resolution for effective dense prediction. In this paper, for more effective guidance of densely encoded features to the desired depth prediction, we propose a network architecture that utilizes novel local planar guidance layers located at multiple stages in the decoding phase. We show that the proposed method outperforms the state-of-the-art works with significant margin evaluating on challenging benchmarks. We also provide results from an ablation study to validate the effectiveness of the proposed method.

In this work, we focus on the task of learning and representing dense correspondences in deformable object categories. While this problem has been considered before, solutions so far have been rather ad-hoc for specific object types (i.e., humans), often with significant manual work involved. However, scaling the geometry understanding to all objects in nature requires more automated approaches that can also express correspondences between related, but geometrically different objects. To this end, we propose a new, learnable image-based representation of dense correspondences. Our model predicts, for each pixel in a 2D image, an embedding vector of the corresponding vertex in the object mesh, therefore establishing dense correspondences between image pixels and 3D object geometry. We demonstrate that the proposed approach performs on par or better than the state-of-the-art methods for dense pose estimation for humans, while being conceptually simpler. We also collect a new in-the-wild dataset of dense correspondences for animal classes and demonstrate that our framework scales naturally to the new deformable object categories.

Visual encoders are fundamental components in vision-language models (VLMs), each showcasing unique strengths derived from various pre-trained visual foundation models. To leverage the various capabilities of these encoders, recent studies incorporate multiple encoders within a single VLM, leading to a considerable increase in computational cost. In this paper, we present Mixture-of-Visual-Encoder Knowledge Distillation (MoVE-KD), a novel framework that distills the unique proficiencies of multiple vision encoders into a single, efficient encoder model. Specifically, to mitigate conflicts and retain the unique characteristics of each teacher encoder, we employ low-rank adaptation (LoRA) and mixture-of-experts (MoEs) to selectively activate specialized knowledge based on input features, enhancing both adaptability and efficiency. To regularize the KD process and enhance performance, we propose an attention-based distillation strategy that adaptively weighs the different visual encoders and emphasizes valuable visual tokens, reducing the burden of replicating comprehensive but distinct features from multiple teachers. Comprehensive experiments on popular VLMs, such as LLaVA and LLaVA-NeXT, validate the effectiveness of our method. The code will be released.

In this work, we establish dense correspondences between RGB image and a surface-based representation of the human body, a task we refer to as dense human pose estimation. We first gather dense correspondences for 50K persons appearing in the COCO dataset by introducing an efficient annotation pipeline. We then use our dataset to train CNN-based systems that deliver dense correspondence 'in the wild', namely in the presence of background, occlusions and scale variations. We improve our training set's effectiveness by training an 'inpainting' network that can fill in missing groundtruth values and report clear improvements with respect to the best results that would be achievable in the past. We experiment with fully-convolutional networks and region-based models and observe a superiority of the latter; we further improve accuracy through cascading, obtaining a system that delivers highly0accurate results in real time. Supplementary materials and videos are provided on the project page http://densepose.org

Existing Multimodal Large Language Models (MLLMs) increasingly emphasize complex understanding of various visual elements, including multiple objects, text information, and spatial relations. Their development for comprehensive visual perception hinges on the availability of high-quality image-text datasets that offer diverse visual elements and throughout image descriptions. However, the scarcity of such hyper-detailed datasets currently hinders progress within the MLLM community. The bottleneck stems from the limited perceptual capabilities of current caption engines, which fall short in providing complete and accurate annotations. To facilitate the cutting-edge research of MLLMs on comprehensive vision perception, we thereby propose Perceptual Fusion, using a low-budget but highly effective caption engine for complete and accurate image descriptions. Specifically, Perceptual Fusion integrates diverse perception experts as image priors to provide explicit information on visual elements and adopts an efficient MLLM as a centric pivot to mimic advanced MLLMs' perception abilities. We carefully select 1M highly representative images from uncurated LAION dataset and generate dense descriptions using our engine, dubbed DenseFusion-1M. Extensive experiments validate that our engine outperforms its counterparts, where the resulting dataset significantly improves the perception and cognition abilities of existing MLLMs across diverse vision-language benchmarks, especially with high-resolution images as inputs. The dataset and code are publicly available at https://github.com/baaivision/DenseFusion.

We propose sorting patch representations across views as a novel self-supervised learning signal to improve pretrained representations. To this end, we introduce NeCo: Patch Neighbor Consistency, a novel training loss that enforces patch-level nearest neighbor consistency across a student and teacher model, relative to reference batches. Our method leverages a differentiable sorting method applied on top of pretrained representations, such as DINOv2-registers to bootstrap the learning signal and further improve upon them. This dense post-pretraining leads to superior performance across various models and datasets, despite requiring only 19 hours on a single GPU. We demonstrate that this method generates high-quality dense feature encoders and establish several new state-of-the-art results: +5.5% and + 6% for non-parametric in-context semantic segmentation on ADE20k and Pascal VOC, and +7.2% and +5.7% for linear segmentation evaluations on COCO-Things and -Stuff.

Several video understanding tasks, such as natural language temporal video grounding, temporal activity localization, and audio description generation, require "temporally dense" reasoning over frames sampled at high temporal resolution. However, computing frame-level features for these tasks is computationally expensive given the temporal resolution requirements. In this paper, we make three contributions to reduce the cost of computing features for temporally dense tasks. First, we introduce a vision transformer (ViT) architecture, dubbed ResidualViT, that leverages the large temporal redundancy in videos to efficiently compute temporally dense frame-level features. Our architecture incorporates (i) learnable residual connections that ensure temporal consistency across consecutive frames and (ii) a token reduction module that enhances processing speed by selectively discarding temporally redundant information while reusing weights of a pretrained foundation model. Second, we propose a lightweight distillation strategy to approximate the frame-level features of the original foundation model. Finally, we evaluate our approach across four tasks and five datasets, in both zero-shot and fully supervised settings, demonstrating significant reductions in computational cost (up to 60%) and improvements in inference speed (up to 2.5x faster), all while closely approximating the accuracy of the original foundation model.

Contrastively-trained Vision-Language Models (VLMs), such as CLIP, have become the standard approach for learning discriminative vision-language representations. However, these models often exhibit shallow language understanding, manifesting bag-of-words behaviour. These limitations are reinforced by their dual-encoder design, which induces a modality gap. Additionally, the reliance on vast web-collected data corpora for training makes the process computationally expensive and introduces significant privacy concerns. To address these limitations, in this work, we challenge the necessity of vision encoders for retrieval tasks by introducing a vision-free, single-encoder retrieval pipeline. Departing from the traditional text-to-image retrieval paradigm, we migrate to a text-to-text paradigm with the assistance of VLLM-generated structured image descriptions. We demonstrate that this paradigm shift has significant advantages, including a substantial reduction of the modality gap, improved compositionality, and better performance on short and long caption queries, all attainable with only a few hours of calibration on two GPUs. Additionally, substituting raw images with textual descriptions introduces a more privacy-friendly alternative for retrieval. To further assess generalisation and address some of the shortcomings of prior compositionality benchmarks, we release two benchmarks derived from Flickr30k and COCO, containing diverse compositional queries made of short captions, which we coin subFlickr and subCOCO. Our vision-free retriever matches and often surpasses traditional multimodal models. Importantly, our approach achieves state-of-the-art zero-shot performance on multiple retrieval and compositionality benchmarks, with models as small as 0.3B parameters. Code is available at: https://github.com/IoannaNti/LexiCLIP

Vision transformers have recently achieved competitive results across various vision tasks but still suffer from heavy computation costs when processing a large number of tokens. Many advanced approaches have been developed to reduce the total number of tokens in large-scale vision transformers, especially for image classification tasks. Typically, they select a small group of essential tokens according to their relevance with the class token, then fine-tune the weights of the vision transformer. Such fine-tuning is less practical for dense prediction due to the much heavier computation and GPU memory cost than image classification. In this paper, we focus on a more challenging problem, i.e., accelerating large-scale vision transformers for dense prediction without any additional re-training or fine-tuning. In response to the fact that high-resolution representations are necessary for dense prediction, we present two non-parametric operators, a token clustering layer to decrease the number of tokens and a token reconstruction layer to increase the number of tokens. The following steps are performed to achieve this: (i) we use the token clustering layer to cluster the neighboring tokens together, resulting in low-resolution representations that maintain the spatial structures; (ii) we apply the following transformer layers only to these low-resolution representations or clustered tokens; and (iii) we use the token reconstruction layer to re-create the high-resolution representations from the refined low-resolution representations. The results obtained by our method are promising on five dense prediction tasks, including object detection, semantic segmentation, panoptic segmentation, instance segmentation, and depth estimation.

Due to the auto-regressive nature of current video large language models (Video-LLMs), the inference latency increases as the input sequence length grows, posing challenges for the efficient processing of video sequences that are usually very long. We observe that during decoding, the attention scores of most tokens in Video-LLMs tend to be sparse and concentrated, with only certain tokens requiring comprehensive full attention. Based on this insight, we introduce Sparse-to-Dense (StD), a novel decoding strategy that integrates two distinct modules: one leveraging sparse top-K attention and the other employing dense full attention. These modules collaborate to accelerate Video-LLMs without loss. The fast (sparse) model speculatively decodes multiple tokens, while the slow (dense) model verifies them in parallel. StD is a tuning-free, plug-and-play solution that achieves up to a 1.94times walltime speedup in video processing. It maintains model performance while enabling a seamless transition from a standard Video-LLM to a sparse Video-LLM with minimal code modifications.

This paper presents a family of advanced vision encoder, named OpenVision 3, that learns a single, unified visual representation that can serve both image understanding and image generation. Our core architecture is simple: we feed VAE-compressed image latents to a ViT encoder and train its output to support two complementary roles. First, the encoder output is passed to the ViT-VAE decoder to reconstruct the original image, encouraging the representation to capture generative structure. Second, the same representation is optimized with contrastive learning and image-captioning objectives, strengthening semantic features. By jointly optimizing reconstruction- and semantics-driven signals in a shared latent space, the encoder learns representations that synergize and generalize well across both regimes. We validate this unified design through extensive downstream evaluations with the encoder frozen. For multimodal understanding, we plug the encoder into the LLaVA-1.5 framework: it performs comparably with a standard CLIP vision encoder (e.g., 62.4 vs 62.2 on SeedBench, and 83.7 vs 82.9 on POPE). For generation, we test it under the RAE framework: ours substantially surpasses the standard CLIP-based encoder (e.g., gFID: 1.89 vs 2.54 on ImageNet). We hope this work can spur future research on unified modeling.

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 https://gdm-tipsv2.github.io/ .

Existing vision-language models (VLMs) mostly rely on vision encoders to extract visual features followed by large language models (LLMs) for visual-language tasks. However, the vision encoders set a strong inductive bias in abstracting visual representation, e.g., resolution, aspect ratio, and semantic priors, which could impede the flexibility and efficiency of the VLMs. Training pure VLMs that accept the seamless vision and language inputs, i.e., without vision encoders, remains challenging and rarely explored. Empirical observations reveal that direct training without encoders results in slow convergence and large performance gaps. In this work, we bridge the gap between encoder-based and encoder-free models, and present a simple yet effective training recipe towards pure VLMs. Specifically, we unveil the key aspects of training encoder-free VLMs efficiently via thorough experiments: (1) Bridging vision-language representation inside one unified decoder; (2) Enhancing visual recognition capability via extra supervision. With these strategies, we launch EVE, an encoder-free vision-language model that can be trained and forwarded efficiently. Notably, solely utilizing 35M publicly accessible data, EVE can impressively rival the encoder-based VLMs of similar capacities across multiple vision-language benchmarks. It significantly outperforms the counterpart Fuyu-8B with mysterious training procedures and undisclosed training data. We believe that EVE provides a transparent and efficient route for developing a pure decoder-only architecture across modalities. Our code and models are publicly available at: https://github.com/baaivision/EVE.

As generative technologies advance, visual content has evolved into a complex mix of natural and AI-generated images, driving the need for more efficient coding techniques that prioritize perceptual quality. Traditional codecs and learned methods struggle to maintain subjective quality at high compression ratios, while existing generative approaches face challenges in visual fidelity and generalization. To this end, we propose a novel generative coding framework leveraging diffusion priors to enhance compression performance at low bitrates. Our approach employs a pre-optimized encoder to generate generalized compressed-domain representations, integrated with the pretrained model's internal features via a lightweight adapter and an attentive fusion module. This framework effectively leverages existing pretrained diffusion models and enables efficient adaptation to different pretrained models for new requirements with minimal retraining costs. We also introduce a distribution renormalization method to further enhance reconstruction fidelity. Extensive experiments show that our method (1) outperforms existing methods in visual fidelity across low bitrates, (2) improves compression performance by up to 79% over H.266/VVC, and (3) offers an efficient solution for AI-generated content while being adaptable to broader content types.

Recent Large Vision-Language Models (LVLMs) have advanced multi-modal understanding by incorporating finer-grained visual perception and encoding. However, such methods incur significant computational costs due to longer visual token sequences, posing challenges for real-time deployment. To mitigate this, prior studies have explored pruning unimportant visual tokens either at the output layer of the visual encoder or at the early layers of the language model. In this work, we revisit these design choices and reassess their effectiveness through comprehensive empirical studies of how visual tokens are processed throughout the visual encoding and language decoding stages. Guided by these insights, we propose VScan, a two-stage visual token reduction framework that addresses token redundancy by: (1) integrating complementary global and local scans with token merging during visual encoding, and (2) introducing pruning at intermediate layers of the language model. Extensive experimental results across four LVLMs validate the effectiveness of VScan in accelerating inference and demonstrate its superior performance over current state-of-the-arts on sixteen benchmarks. Notably, when applied to LLaVA-NeXT-7B, VScan achieves a 2.91times speedup in prefilling and a 10times reduction in FLOPs, while retaining 95.4% of the original performance.

Representation autoencoders that reuse frozen pretrained vision encoders as visual tokenizers have achieved strong reconstruction and generation quality. However, existing methods universally extract features from only the last encoder layer, discarding the rich hierarchical information distributed across intermediate layers. We show that low-level visual details survive in the last layer merely as attenuated residuals after multiple layers of semantic abstraction, and that explicitly fusing multi-layer features can substantially recover this lost information. We propose DRoRAE (Depth-Routed Representation AutoEncoder), a lightweight fusion module that adaptively aggregates all encoder layers via energy-constrained routing and incremental correction, producing an enriched latent compatible with a frozen pretrained decoder. A three-phase decoupled training strategy first learns the fusion under the implicit distributional constraint of the frozen decoder, then fine-tunes the decoder to fully exploit the enriched representation. On ImageNet-256, DRoRAE reduces rFID from 0.57 to 0.29 and improves generation FID from 1.74 to 1.65 (with AutoGuidance), with gains also transferring to text-to-image synthesis. Furthermore, we uncover a log-linear scaling law (R^2{=}0.86) between fusion capacity and reconstruction quality, identifying representation richness as a new, predictably scalable dimension for visual tokenizers analogous to vocabulary size in NLP.

The space of task-agnostic feature upsampling has emerged as a promising area of research to efficiently create denser features from pre-trained visual backbones. These methods act as a shortcut to achieve dense features for a fraction of the cost by learning to map low-resolution features to high-resolution versions. While early works in this space used iterative upsampling approaches, more recent works have switched to cross-attention-based methods, which risk falling into the same efficiency scaling problems of the backbones they are upsampling. In this work, we demonstrate that iterative upsampling methods can still compete with cross-attention-based methods; moreover, they can achieve state-of-the-art performance with lower inference costs. We propose UPLiFT, an architecture for Universal Pixel-dense Lightweight Feature Transforms. We also propose an efficient Local Attender operator to overcome the limitations of prior iterative feature upsampling methods. This operator uses an alternative attentional pooling formulation defined fully locally. We show that our Local Attender allows UPLiFT to maintain stable features throughout upsampling, enabling state-of-the-art performance with lower inference costs than existing pixel-dense feature upsamplers. In addition, we apply UPLiFT to generative downstream tasks and show that it achieves competitive performance with state-of-the-art Coupled Flow Matching models for VAE feature upsampling. Altogether, UPLiFT offers a versatile and efficient approach to creating denser features.

In this paper, we introduce a novel visual representation learning which relies on a handful of adaptively learned tokens, and which is applicable to both image and video understanding tasks. Instead of relying on hand-designed splitting strategies to obtain visual tokens and processing a large number of densely sampled patches for attention, our approach learns to mine important tokens in visual data. This results in efficiently and effectively finding a few important visual tokens and enables modeling of pairwise attention between such tokens, over a longer temporal horizon for videos, or the spatial content in images. Our experiments demonstrate strong performance on several challenging benchmarks for both image and video recognition tasks. Importantly, due to our tokens being adaptive, we accomplish competitive results at significantly reduced compute amount. We obtain comparable results to the state-of-the-arts on ImageNet while being computationally more efficient. We also confirm the effectiveness of the approach on multiple video datasets, including Kinetics-400, Kinetics-600, Charades, and AViD. The code is available at: https://github.com/google-research/scenic/tree/main/scenic/projects/token_learner

Multimodal Large Language Models (MLLMs) have recently achieved remarkable success in visual-language understanding, demonstrating superior high-level semantic alignment within their vision encoders. An important question thus arises: Can these encoders serve as versatile vision backbones, capable of reliably performing classic vision-centric tasks as well? To address the question, we make the following contributions: (i) we identify that the vision encoders within MLLMs exhibit deficiencies in their dense feature representations, as evidenced by their suboptimal performance on dense prediction tasks (e.g., semantic segmentation, depth estimation); (ii) we propose VersaViT, a well-rounded vision transformer that instantiates a novel multi-task framework for collaborative post-training. This framework facilitates the optimization of the vision backbone via lightweight task heads with multi-granularity supervision; (iii) extensive experiments across various downstream tasks demonstrate the effectiveness of our method, yielding a versatile vision backbone suited for both language-mediated reasoning and pixel-level understanding.

Self-supervised learning holds the promise of eliminating the need for manual data annotation, enabling models to scale effortlessly to massive datasets and larger architectures. By not being tailored to specific tasks or domains, this training paradigm has the potential to learn visual representations from diverse sources, ranging from natural to aerial images -- using a single algorithm. This technical report introduces DINOv3, a major milestone toward realizing this vision by leveraging simple yet effective strategies. First, we leverage the benefit of scaling both dataset and model size by careful data preparation, design, and optimization. Second, we introduce a new method called Gram anchoring, which effectively addresses the known yet unsolved issue of dense feature maps degrading during long training schedules. Finally, we apply post-hoc strategies that further enhance our models' flexibility with respect to resolution, model size, and alignment with text. As a result, we present a versatile vision foundation model that outperforms the specialized state of the art across a broad range of settings, without fine-tuning. DINOv3 produces high-quality dense features that achieve outstanding performance on various vision tasks, significantly surpassing previous self- and weakly-supervised foundation models. We also share the DINOv3 suite of vision models, designed to advance the state of the art on a wide spectrum of tasks and data by providing scalable solutions for diverse resource constraints and deployment scenarios.

We present an unsupervised visual feature learning algorithm driven by context-based pixel prediction. By analogy with auto-encoders, we propose Context Encoders -- a convolutional neural network trained to generate the contents of an arbitrary image region conditioned on its surroundings. In order to succeed at this task, context encoders need to both understand the content of the entire image, as well as produce a plausible hypothesis for the missing part(s). When training context encoders, we have experimented with both a standard pixel-wise reconstruction loss, as well as a reconstruction plus an adversarial loss. The latter produces much sharper results because it can better handle multiple modes in the output. We found that a context encoder learns a representation that captures not just appearance but also the semantics of visual structures. We quantitatively demonstrate the effectiveness of our learned features for CNN pre-training on classification, detection, and segmentation tasks. Furthermore, context encoders can be used for semantic inpainting tasks, either stand-alone or as initialization for non-parametric methods.

To date, most existing self-supervised learning methods are designed and optimized for image classification. These pre-trained models can be sub-optimal for dense prediction tasks due to the discrepancy between image-level prediction and pixel-level prediction. To fill this gap, we aim to design an effective, dense self-supervised learning method that directly works at the level of pixels (or local features) by taking into account the correspondence between local features. We present dense contrastive learning, which implements self-supervised learning by optimizing a pairwise contrastive (dis)similarity loss at the pixel level between two views of input images. Compared to the baseline method MoCo-v2, our method introduces negligible computation overhead (only <1% slower), but demonstrates consistently superior performance when transferring to downstream dense prediction tasks including object detection, semantic segmentation and instance segmentation; and outperforms the state-of-the-art methods by a large margin. Specifically, over the strong MoCo-v2 baseline, our method achieves significant improvements of 2.0% AP on PASCAL VOC object detection, 1.1% AP on COCO object detection, 0.9% AP on COCO instance segmentation, 3.0% mIoU on PASCAL VOC semantic segmentation and 1.8% mIoU on Cityscapes semantic segmentation. Code is available at: https://git.io/AdelaiDet

Object hallucination in Large Vision-Language Models (LVLMs) significantly impedes their real-world applicability. As the primary component for accurately interpreting visual information, the choice of visual encoder is pivotal. We hypothesize that the diverse training paradigms employed by different visual encoders instill them with distinct inductive biases, which leads to their diverse hallucination performances. Existing benchmarks typically focus on coarse-grained hallucination detection and fail to capture the diverse hallucinations elaborated in our hypothesis. To systematically analyze these effects, we introduce VHBench-10, a comprehensive benchmark with approximately 10,000 samples for evaluating LVLMs across ten fine-grained hallucination categories. Our evaluations confirm encoders exhibit unique hallucination characteristics. Building on these insights and the suboptimality of simple feature fusion, we propose VisionWeaver, a novel Context-Aware Routing Network. It employs global visual features to generate routing signals, dynamically aggregating visual features from multiple specialized experts. Comprehensive experiments confirm the effectiveness of VisionWeaver in significantly reducing hallucinations and improving overall model performance.

We present a novel method for local image feature matching. Instead of performing image feature detection, description, and matching sequentially, we propose to first establish pixel-wise dense matches at a coarse level and later refine the good matches at a fine level. In contrast to dense methods that use a cost volume to search correspondences, we use self and cross attention layers in Transformer to obtain feature descriptors that are conditioned on both images. The global receptive field provided by Transformer enables our method to produce dense matches in low-texture areas, where feature detectors usually struggle to produce repeatable interest points. The experiments on indoor and outdoor datasets show that LoFTR outperforms state-of-the-art methods by a large margin. LoFTR also ranks first on two public benchmarks of visual localization among the published methods.

Dense prediction tasks hold significant importance of computer vision, aiming to learn pixel-wise annotated label for an input image. Despite advances in this field, existing methods primarily focus on idealized conditions, with limited generalization to real-world scenarios and facing the challenging scarcity of real-world data. To systematically study this problem, we first introduce DenseWorld, a benchmark spanning a broad set of 25 dense prediction tasks that correspond to urgent real-world applications, featuring unified evaluation across tasks. Then, we propose DenseDiT, which maximally exploits generative models' visual priors to perform diverse real-world dense prediction tasks through a unified strategy. DenseDiT combines a parameter-reuse mechanism and two lightweight branches that adaptively integrate multi-scale context, working with less than 0.1% additional parameters. Evaluations on DenseWorld reveal significant performance drops in existing general and specialized baselines, highlighting their limited real-world generalization. In contrast, DenseDiT achieves superior results using less than 0.01% training data of baselines, underscoring its practical value for real-world deployment. Our data, and checkpoints and codes are available at https://xcltql666.github.io/DenseDiTProj

Running AI models on smart edge devices can unlock versatile user experiences, but presents challenges due to limited compute and the need to handle multiple tasks simultaneously. This requires a vision encoder with small size but powerful and versatile representations. We present our method, Efficient Universal Perception Encoder (EUPE), which offers both inference efficiency and universally good representations for diverse downstream tasks. We achieve this by distilling from multiple domain-expert foundation vision encoders. Unlike previous agglomerative methods that directly scale down from multiple teachers to an efficient encoder, we demonstrate the importance of first scaling up to a large proxy teacher and then scaling down from this single teacher. Experiments show that EUPE achieves on-par or better performance than individual domain experts of the same size on diverse task domains and also outperforms previous agglomerative encoders. We will release the full family of EUPE models and the code to foster future research.

Vision transformers have been widely explored in various vision tasks. Due to heavy computational cost, much interest has aroused for compressing vision transformer dynamically in the aspect of tokens. Current methods mainly pay attention to token pruning or merging to reduce token numbers, in which tokens are compressed exclusively, causing great information loss and therefore post-training is inevitably required to recover the performance. In this paper, we rethink token reduction and unify the process as an explicit form of token matrix transformation, in which all existing methods are constructing special forms of matrices within the framework. Furthermore, we propose a many-to-many Token Transforming framework that serves as a generalization of all existing methods and reserves the most information, even enabling training-free acceleration. We conduct extensive experiments to validate our framework. Specifically, we reduce 40% FLOPs and accelerate DeiT-S by times1.5 with marginal 0.1% accuracy drop. Furthermore, we extend the method to dense prediction tasks including segmentation, object detection, depth estimation, and language model generation. Results demonstrate that the proposed method consistently achieves substantial improvements, offering a better computation-performance trade-off, impressive budget reduction and inference acceleration.

Despite the great success of Large Vision-Language Models (LVLMs), they inevitably suffer from hallucination. As we know, both the visual encoder and the Large Language Model (LLM) decoder in LVLMs are Transformer-based, allowing the model to extract visual information and generate text outputs via attention mechanisms. We find that the attention distribution of LLM decoder on image tokens is highly consistent with the visual encoder and both distributions tend to focus on particular background tokens rather than the referred objects in the image. We attribute to the unexpected attention distribution to an inherent flaw in the visual encoder itself, which misguides LLMs to over emphasize the redundant information and generate object hallucination. To address the issue, we propose DAMRO, a novel training-free strategy that Dive into Attention Mechanism of LVLM to Reduce Object Hallucination. Specifically, our approach employs classification token (CLS) of ViT to filter out high-attention outlier tokens scattered in the background and then eliminate their influence during decoding stage. We evaluate our method on LVLMs including LLaVA-1.5, LLaVA-NeXT and InstructBLIP, using various benchmarks such as POPE, CHAIR, MME and GPT-4V Aided Evaluation. The results demonstrate that our approach significantly reduces the impact of these outlier tokens, thus effectively alleviating the hallucination of LVLMs. The code of our method will be released soon.

The application of Large Vision-Language Models (LVLMs) for analyzing images and videos is an exciting and rapidly evolving field. In recent years, we've seen significant growth in high-quality image-text datasets for fine-tuning image understanding, but there is still a lack of comparable datasets for videos. Additionally, many VideoLLMs are extensions of single-image VLMs, which may not efficiently handle the complexities of longer videos. In this study, we introduce a large-scale synthetic dataset created from proprietary models, using carefully designed prompts to tackle a wide range of questions. We also explore a dynamic visual token compression architecture that strikes a balance between computational efficiency and performance. Our proposed achieves state-of-the-art results across various video tasks and shows impressive generalization, setting new baselines in multi-image understanding. Notably, delivers an absolute improvement of 2.7\% over LLaVA-OneVision on VideoMME and 10.7\% on MuirBench. Codes are available at https://github.com/Hon-Wong/ByteVideoLLM

Dual encoders perform retrieval by encoding documents and queries into dense lowdimensional vectors, scoring each document by its inner product with the query. We investigate the capacity of this architecture relative to sparse bag-of-words models and attentional neural networks. Using both theoretical and empirical analysis, we establish connections between the encoding dimension, the margin between gold and lower-ranked documents, and the document length, suggesting limitations in the capacity of fixed-length encodings to support precise retrieval of long documents. Building on these insights, we propose a simple neural model that combines the efficiency of dual encoders with some of the expressiveness of more costly attentional architectures, and explore sparse-dense hybrids to capitalize on the precision of sparse retrieval. These models outperform strong alternatives in large-scale retrieval.

Vision encoders typically generate a large number of visual tokens, providing information-rich representations but significantly increasing computational demands. This raises the question of whether all generated tokens are equally valuable or if some of them can be discarded to reduce computational costs without compromising quality. In this paper, we introduce a new method for determining feature utility based on the idea that less valuable features can be reconstructed from more valuable ones. We implement this concept by integrating an autoencoder with a Gumbel-Softmax selection mechanism, that allows identifying and retaining only the most informative visual tokens. To validate our approach, we compared the performance of the LLaVA-NeXT model, using features selected by our method with randomly selected features. We found that on OCR-based tasks, more than 50% of the visual context can be removed with minimal performance loss, whereas randomly discarding the same proportion of features significantly affects the model capabilities. Furthermore, in general-domain tasks, even randomly retaining only 30% of tokens achieves performance comparable to using the full set of visual tokens. Our results highlight a promising direction towards adaptive and efficient multimodal pruning that facilitates scalable and low-overhead inference without compromising performance.

Computer vision has achieved remarkable success by (a) representing images as uniformly-arranged pixel arrays and (b) convolving highly-localized features. However, convolutions treat all image pixels equally regardless of importance; explicitly model all concepts across all images, regardless of content; and struggle to relate spatially-distant concepts. In this work, we challenge this paradigm by (a) representing images as semantic visual tokens and (b) running transformers to densely model token relationships. Critically, our Visual Transformer operates in a semantic token space, judiciously attending to different image parts based on context. This is in sharp contrast to pixel-space transformers that require orders-of-magnitude more compute. Using an advanced training recipe, our VTs significantly outperform their convolutional counterparts, raising ResNet accuracy on ImageNet top-1 by 4.6 to 7 points while using fewer FLOPs and parameters. For semantic segmentation on LIP and COCO-stuff, VT-based feature pyramid networks (FPN) achieve 0.35 points higher mIoU while reducing the FPN module's FLOPs by 6.5x.

Vision encoders are increasingly used in modern applications, from vision-only models to multimodal systems such as vision-language models. Despite their remarkable success, it remains unclear how these architectures represent features internally. Here, we propose a novel approach for interpreting vision features via image reconstruction. We compare two related model families, SigLIP and SigLIP2, which differ only in their training objective, and show that encoders pre-trained on image-based tasks retain significantly more image information than those trained on non-image tasks such as contrastive learning. We further apply our method to a range of vision encoders, ranking them by the informativeness of their feature representations. Finally, we demonstrate that manipulating the feature space yields predictable changes in reconstructed images, revealing that orthogonal rotations (rather than spatial transformations) control color encoding. Our approach can be applied to any vision encoder, shedding light on the inner structure of its feature space. The code and model weights to reproduce the experiments are available in GitHub.

Enabling Visual Semantic Models to effectively handle multi-view description matching has been a longstanding challenge. Existing methods typically learn a set of embeddings to find the optimal match for each view's text and compute similarity. However, the visual and text embeddings learned through these approaches have limited information capacity and are prone to interference from locally similar negative samples. To address this issue, we argue that the information capacity of embeddings is crucial and propose Dense-to-Sparse Feature Distilled Visual Semantic Embedding (D2S-VSE), which enhances the information capacity of sparse text by leveraging dense text distillation. Specifically, D2S-VSE is a two-stage framework. In the pre-training stage, we align images with dense text to enhance the information capacity of visual semantic embeddings. In the fine-tuning stage, we optimize two tasks simultaneously, distilling dense text embeddings to sparse text embeddings while aligning images and sparse texts, enhancing the information capacity of sparse text embeddings. Our proposed D2S-VSE model is extensively evaluated on the large-scale MS-COCO and Flickr30K datasets, demonstrating its superiority over recent state-of-the-art methods.

The recently proposed SparseFormer architecture provides an alternative approach to visual understanding by utilizing a significantly lower number of visual tokens via adjusting RoIs, greatly reducing computational costs while still achieving promising performance. However, training SparseFormers from scratch is still expensive, and scaling up the number of parameters can be challenging. In this paper, we propose to bootstrap SparseFormers from ViT-based vision foundation models in a simple and efficient way. Since the majority of SparseFormer blocks are the standard transformer ones, we can inherit weights from large-scale pre-trained vision transformers and freeze them as much as possible. Therefore, we only need to train the SparseFormer-specific lightweight focusing transformer to adjust token RoIs and fine-tune a few early pre-trained blocks to align the final token representation. In such a way, we can bootstrap SparseFormer architectures from various large-scale pre-trained models (e.g., IN-21K pre-trained AugRegs or CLIPs) using a rather smaller amount of training samples (e.g., IN-1K) and without labels or captions within just a few hours. As a result, the bootstrapped unimodal SparseFormer (from AugReg-ViT-L/16-384) can reach 84.9% accuracy on IN-1K with only 49 tokens, and the multimodal SparseFormer from CLIPs also demonstrates notable zero-shot performance with highly reduced computational cost without seeing any caption during the bootstrapping procedure. In addition, CLIP-bootstrapped SparseFormers, which align the output space with language without seeing a word, can serve as efficient vision encoders in multimodal large language models. Code will be publicly available at https://github.com/showlab/sparseformer

In this work, we focus on the challenge of temporally consistent human-centric dense prediction across video sequences. Existing models achieve strong per-frame accuracy but often flicker under motion, occlusion, and lighting changes, and they rarely have paired human video supervision for multiple dense tasks. We address this gap with a scalable synthetic data pipeline that generates photorealistic human frames and motion-aligned sequences with pixel-accurate depth, normals, and masks. Unlike prior static data synthetic pipelines, our pipeline provides both frame-level labels for spatial learning and sequence-level supervision for temporal learning. Building on this, we train a unified ViT-based dense predictor that (i) injects an explicit human geometric prior via CSE embeddings and (ii) improves geometry-feature reliability with a lightweight channel reweighting module after feature fusion. Our two-stage training strategy, combining static pretraining with dynamic sequence supervision, enables the model first to acquire robust spatial representations and then refine temporal consistency across motion-aligned sequences. Extensive experiments show that we achieve state-of-the-art performance on THuman2.1 and Hi4D and generalize effectively to in-the-wild videos.

Vision-language models (VLMs) benefit from multiple vision encoders, but naively stacking them yields diminishing returns while multiplying inference costs. We propose SCOPE, a Mixture-of-Encoders (MoEnc) framework that dynamically selects one specialized encoder per image-text pair via instance-level routing, unlike token-level routing in traditional MoE. SCOPE maintains a shared encoder and a pool of routed encoders. A lightweight router uses cross-attention between text prompts and shared visual features to select the optimal encoder from the routed encoders. To train this router, we introduce dual entropy regularization with auxiliary losses to balance dataset-level load distribution with instance-level routing confidence. Remarkably, SCOPE with one shared plus one routed encoder outperforms models using all four extra encoders simultaneously, while reducing compute by 24-49\%. This demonstrates that intelligent encoder selection beats brute-force aggregation, challenging the prevailing paradigm in multi-encoder VLMs.

This work investigates a simple yet powerful dense prediction task adapter for Vision Transformer (ViT). Unlike recently advanced variants that incorporate vision-specific inductive biases into their architectures, the plain ViT suffers inferior performance on dense predictions due to weak prior assumptions. To address this issue, we propose the ViT-Adapter, which allows plain ViT to achieve comparable performance to vision-specific transformers. Specifically, the backbone in our framework is a plain ViT that can learn powerful representations from large-scale multi-modal data. When transferring to downstream tasks, a pre-training-free adapter is used to introduce the image-related inductive biases into the model, making it suitable for these tasks. We verify ViT-Adapter on multiple dense prediction tasks, including object detection, instance segmentation, and semantic segmentation. Notably, without using extra detection data, our ViT-Adapter-L yields state-of-the-art 60.9 box AP and 53.0 mask AP on COCO test-dev. We hope that the ViT-Adapter could serve as an alternative for vision-specific transformers and facilitate future research. The code and models will be released at https://github.com/czczup/ViT-Adapter.

Dense prediction tasks, such as semantic segmentation, depth estimation, and surface normal prediction, can be easily formulated as per-pixel classification (discrete outputs) or regression (continuous outputs). This per-pixel prediction paradigm has remained popular due to the prevalence of fully convolutional networks. However, on the recent frontier of segmentation task, the community has been witnessing a shift of paradigm from per-pixel prediction to cluster-prediction with the emergence of transformer architectures, particularly the mask transformers, which directly predicts a label for a mask instead of a pixel. Despite this shift, methods based on the per-pixel prediction paradigm still dominate the benchmarks on the other dense prediction tasks that require continuous outputs, such as depth estimation and surface normal prediction. Motivated by the success of DORN and AdaBins in depth estimation, achieved by discretizing the continuous output space, we propose to generalize the cluster-prediction based method to general dense prediction tasks. This allows us to unify dense prediction tasks with the mask transformer framework. Remarkably, the resulting model PolyMaX demonstrates state-of-the-art performance on three benchmarks of NYUD-v2 dataset. We hope our simple yet effective design can inspire more research on exploiting mask transformers for more dense prediction tasks. Code and model will be made available.

We present V-JEPA 2.1, a family of self-supervised models that learn dense, high-quality visual representations for both images and videos while retaining strong global scene understanding. The approach combines four key components. First, a dense predictive loss uses a masking-based objective in which both visible and masked tokens contribute to the training signal, encouraging explicit spatial and temporal grounding. Second, deep self-supervision applies the self-supervised objective hierarchically across multiple intermediate encoder layers to improve representation quality. Third, multi-modal tokenizers enable unified training across images and videos. Finally, the model benefits from effective scaling in both model capacity and training data. Together, these design choices produce representations that are spatially structured, semantically coherent, and temporally consistent. Empirically, V-JEPA 2.1 achieves state-of-the-art performance on several challenging benchmarks, including 7.71 mAP on Ego4D for short-term object-interaction anticipation and 40.8 Recall@5 on EPIC-KITCHENS for high-level action anticipation, as well as a 20-point improvement in real-robot grasping success rate over V-JEPA-2 AC. The model also demonstrates strong performance in robotic navigation (5.687 ATE on TartanDrive), depth estimation (0.307 RMSE on NYUv2 with a linear probe), and global recognition (77.7 on Something-Something-V2). These results show that V-JEPA 2.1 significantly advances the state of the art in dense visual understanding and world modeling.

We present Multiscale Vision Transformers (MViT) for video and image recognition, by connecting the seminal idea of multiscale feature hierarchies with transformer models. Multiscale Transformers have several channel-resolution scale stages. Starting from the input resolution and a small channel dimension, the stages hierarchically expand the channel capacity while reducing the spatial resolution. This creates a multiscale pyramid of features with early layers operating at high spatial resolution to model simple low-level visual information, and deeper layers at spatially coarse, but complex, high-dimensional features. We evaluate this fundamental architectural prior for modeling the dense nature of visual signals for a variety of video recognition tasks where it outperforms concurrent vision transformers that rely on large scale external pre-training and are 5-10x more costly in computation and parameters. We further remove the temporal dimension and apply our model for image classification where it outperforms prior work on vision transformers. Code is available at: https://github.com/facebookresearch/SlowFast

We present SOLO, a single transformer for Scalable visiOn-Language mOdeling. Current large vision-language models (LVLMs) such as LLaVA mostly employ heterogeneous architectures that connect pre-trained visual encoders with large language models (LLMs) to facilitate visual recognition and complex reasoning. Although achieving remarkable performance with relatively lightweight training, we identify four primary scalability limitations: (1) The visual capacity is constrained by pre-trained visual encoders, which are typically an order of magnitude smaller than LLMs. (2) The heterogeneous architecture complicates the use of established hardware and software infrastructure. (3) Study of scaling laws on such architecture must consider three separate components - visual encoder, connector, and LLMs, which complicates the analysis. (4) The use of existing visual encoders typically requires following a pre-defined specification of image inputs pre-processing, for example, by reshaping inputs to fixed-resolution square images, which presents difficulties in processing and training on high-resolution images or those with unusual aspect ratio. A unified single Transformer architecture, like SOLO, effectively addresses these scalability concerns in LVLMs; however, its limited adoption in the modern context likely stems from the absence of reliable training recipes that balance both modalities and ensure stable training for billion-scale models. In this paper, we introduce the first open-source training recipe for developing SOLO, an open-source 7B LVLM using moderate academic resources. The training recipe involves initializing from LLMs, sequential pre-training on ImageNet and web-scale data, and instruction fine-tuning on our curated high-quality datasets. On extensive evaluation, SOLO demonstrates performance comparable to LLaVA-v1.5-7B, particularly excelling in visual mathematical reasoning.

We propose a new task and model for dense video object captioning -- detecting, tracking and captioning trajectories of objects in a video. This task unifies spatial and temporal localization in video, whilst also requiring fine-grained visual understanding that is best described by natural language. We propose a unified model, and demonstrate how our end-to-end approach is more accurate and temporally coherent than a multi-stage pipeline combining state-of-the-art detection, tracking, and captioning models. Moreover, we propose a training strategy based on a mixture of disjoint tasks, which allows us to leverage diverse, large-scale datasets which supervise different parts of our model. Although each pretraining task only provides weak supervision, they are complementary and, when combined, result in noteworthy zero-shot ability and serve as strong initialization for additional finetuning to further improve accuracy. We carefully design new metrics capturing all components of our task, and show how we can repurpose existing video grounding datasets (e.g. VidSTG and VLN) for our new task. We show that our model improves upon a number of strong baselines for this new task. Furthermore, we can apply our model to the task of spatial grounding, outperforming prior state-of-the-art on VidSTG and VLN, without explicitly training for it. Code is available at https://github.com/google-research/scenic/tree/main/scenic/projects/densevoc.

We launch EVA, a vision-centric foundation model to explore the limits of visual representation at scale using only publicly accessible data. EVA is a vanilla ViT pre-trained to reconstruct the masked out image-text aligned vision features conditioned on visible image patches. Via this pretext task, we can efficiently scale up EVA to one billion parameters, and sets new records on a broad range of representative vision downstream tasks, such as image recognition, video action recognition, object detection, instance segmentation and semantic segmentation without heavy supervised training. Moreover, we observe quantitative changes in scaling EVA result in qualitative changes in transfer learning performance that are not present in other models. For instance, EVA takes a great leap in the challenging large vocabulary instance segmentation task: our model achieves almost the same state-of-the-art performance on LVISv1.0 dataset with over a thousand categories and COCO dataset with only eighty categories. Beyond a pure vision encoder, EVA can also serve as a vision-centric, multi-modal pivot to connect images and text. We find initializing the vision tower of a giant CLIP from EVA can greatly stabilize the training and outperform the training from scratch counterpart with much fewer samples and less compute, providing a new direction for scaling up and accelerating the costly training of multi-modal foundation models. To facilitate future research, we release all the code and models at https://github.com/baaivision/EVA.

The exponential growth of Large Multimodal Models (LMMs) has driven advancements in cross-modal reasoning but at significant computational costs. In this work, we focus on visual language models. We highlight the redundancy and inefficiency in current vision encoders, and seek to construct an adaptive compression method for multimodal data. In this work, we characterize a panoply of visual token selection and merging approaches through both benchmarking and qualitative analysis. In particular, we demonstrate that simple cluster-level token aggregation outperforms prior state-of-the-art works in token selection and merging, including merging at the vision encoder level and attention-based approaches. We underline the redundancy in current vision encoders, and shed light on several puzzling trends regarding principles of visual token selection through cross-modal attention visualizations. This work is a first effort towards more effective encoding and processing of high-dimensional data, and paves the way for more scalable and sustainable multimodal systems.

Recent progress has shown that large-scale pre-training using contrastive image-text pairs can be a promising alternative for high-quality visual representation learning from natural language supervision. Benefiting from a broader source of supervision, this new paradigm exhibits impressive transferability to downstream classification tasks and datasets. However, the problem of transferring the knowledge learned from image-text pairs to more complex dense prediction tasks has barely been visited. In this work, we present a new framework for dense prediction by implicitly and explicitly leveraging the pre-trained knowledge from CLIP. Specifically, we convert the original image-text matching problem in CLIP to a pixel-text matching problem and use the pixel-text score maps to guide the learning of dense prediction models. By further using the contextual information from the image to prompt the language model, we are able to facilitate our model to better exploit the pre-trained knowledge. Our method is model-agnostic, which can be applied to arbitrary dense prediction systems and various pre-trained visual backbones including both CLIP models and ImageNet pre-trained models. Extensive experiments demonstrate the superior performance of our methods on semantic segmentation, object detection, and instance segmentation tasks. Code is available at https://github.com/raoyongming/DenseCLIP

We present an efficient encoder-free approach for video-language understanding that achieves competitive performance while significantly reducing computational overhead. Current video-language models typically rely on heavyweight image encoders (300M-1.1B parameters) or video encoders (1B-1.4B parameters), creating a substantial computational burden when processing multi-frame videos. Our method introduces a novel Spatio-Temporal Alignment Block (STAB) that directly processes video inputs without requiring pre-trained encoders while using only 45M parameters for visual processing - at least a 6.5times reduction compared to traditional approaches. The STAB architecture combines Local Spatio-Temporal Encoding for fine-grained feature extraction, efficient spatial downsampling through learned attention and separate mechanisms for modeling frame-level and video-level relationships. Our model achieves comparable or superior performance to encoder-based approaches for open-ended video question answering on standard benchmarks. The fine-grained video question-answering evaluation demonstrates our model's effectiveness, outperforming the encoder-based approaches Video-ChatGPT and Video-LLaVA in key aspects like correctness and temporal understanding. Extensive ablation studies validate our architectural choices and demonstrate the effectiveness of our spatio-temporal modeling approach while achieving 3-4times faster processing speeds than previous methods. Code is available at https://github.com/jh-yi/Video-Panda.

While image-text representation learning has become very popular in recent years, existing models tend to lack spatial awareness and have limited direct applicability for dense understanding tasks. For this reason, self-supervised image-only pretraining is still the go-to method for many dense vision applications (e.g. depth estimation, semantic segmentation), despite the lack of explicit supervisory signals. In this paper, we close this gap between image-text and self-supervised learning, by proposing a novel general-purpose image-text model, which can be effectively used off-the-shelf for dense and global vision tasks. Our method, which we refer to as Text-Image Pretraining with Spatial awareness (TIPS), leverages two simple and effective insights. First, on textual supervision: we reveal that replacing noisy web image captions by synthetically generated textual descriptions boosts dense understanding performance significantly, due to a much richer signal for learning spatially aware representations. We propose an adapted training method that combines noisy and synthetic captions, resulting in improvements across both dense and global understanding tasks. Second, on the learning technique: we propose to combine contrastive image-text learning with self-supervised masked image modeling, to encourage spatial coherence, unlocking substantial enhancements for downstream applications. Building on these two ideas, we scale our model using the transformer architecture, trained on a curated set of public images. Our experiments are conducted on 8 tasks involving 16 datasets in total, demonstrating strong off-the-shelf performance on both dense and global understanding, for several image-only and image-text tasks.

Recent advancements in large video-language models have revolutionized video understanding tasks. However, their efficiency is significantly constrained by processing high volumes of visual tokens. Existing token compression strategies apply a fixed compression ratio, ignoring the variability in semantic density among different video clips. Consequently, this lead to inadequate representation of information-rich clips due to insufficient tokens and unnecessary computation on static or content-poor ones. To address this, we propose LangDC, a Language-aware Dynamic Token Compressor. LangDC leverages a lightweight language model to describe video clips, converting them into soft caption tokens as visual representations. Trained with our proposed semantic density-aware supervision, LangDC aims to 1) cover key visual cues necessary for downstream task reasoning and 2) dynamically adjust compression ratios based on scene richness, reflected by descriptions length. Our design mimics how humans dynamically express what they see: complex scenes (seeing more) elicit more detailed language to convey nuances (saying more), whereas simpler scenes are described with fewer words. Experimental results show that our method reduces FLOPs by 49% compared to VideoGPT+ while maintaining competitive performance. Furthermore, qualitative results demonstrate our approach adaptively adjusts the token compression ratio based on video segment richness.

While recent neural codecs achieve strong performance at low bitrates when optimized for perceptual quality, their effectiveness deteriorates significantly under ultra-low bitrate conditions. To mitigate this, generative compression methods leveraging semantic priors from pretrained models have emerged as a promising paradigm. However, existing approaches are fundamentally constrained by a tradeoff between semantic faithfulness and perceptual realism. Methods based on explicit representations preserve content structure but often lack fine-grained textures, whereas implicit methods can synthesize visually plausible details at the cost of semantic drift. In this work, we propose a unified framework that bridges this gap by coherently integrating explicit and implicit representations in a training-free manner. Specifically, We condition a diffusion model on explicit high-level semantics while employing reverse-channel coding to implicitly convey fine-grained details. Moreover, we introduce a plug-in encoder that enables flexible control of the distortion-perception tradeoff by modulating the implicit information. Extensive experiments demonstrate that the proposed framework achieves state-of-the-art rate-perception performance, outperforming existing methods and surpassing DiffC by 29.92%, 19.33%, and 20.89% in DISTS BD-Rate on the Kodak, DIV2K, and CLIC2020 datasets, respectively.

Curation methods for massive vision-language datasets trade off between dataset size and quality. However, even the highest quality of available curated captions are far too short to capture the rich visual detail in an image. To show the value of dense and highly-aligned image-text pairs, we collect the Densely Captioned Images (DCI) dataset, containing 8012 natural images human-annotated with mask-aligned descriptions averaging above 1000 words each. With precise and reliable captions associated with specific parts of an image, we can evaluate vision-language models' (VLMs) understanding of image content with a novel task that matches each caption with its corresponding subcrop. As current models are often limited to 77 text tokens, we also introduce a summarized version (sDCI) in which each caption length is limited. We show that modern techniques that make progress on standard benchmarks do not correspond with significant improvement on our sDCI based benchmark. Lastly, we finetune CLIP using sDCI and show significant improvements over the baseline despite a small training set. By releasing the first human annotated dense image captioning dataset, we hope to enable the development of new benchmarks or fine-tuning recipes for the next generation of VLMs to come.

We present Visual Lexicon, a novel visual language that encodes rich image information into the text space of vocabulary tokens while retaining intricate visual details that are often challenging to convey in natural language. Unlike traditional methods that prioritize either high-level semantics (e.g., CLIP) or pixel-level reconstruction (e.g., VAE), ViLex simultaneously captures rich semantic content and fine visual details, enabling high-quality image generation and comprehensive visual scene understanding. Through a self-supervised learning pipeline, ViLex generates tokens optimized for reconstructing input images using a frozen text-to-image (T2I) diffusion model, preserving the detailed information necessary for high-fidelity semantic-level reconstruction. As an image embedding in the language space, ViLex tokens leverage the compositionality of natural languages, allowing them to be used independently as "text tokens" or combined with natural language tokens to prompt pretrained T2I models with both visual and textual inputs, mirroring how we interact with vision-language models (VLMs). Experiments demonstrate that ViLex achieves higher fidelity in image reconstruction compared to text embeddings--even with a single ViLex token. Moreover, ViLex successfully performs various DreamBooth tasks in a zero-shot, unsupervised manner without fine-tuning T2I models. Additionally, ViLex serves as a powerful vision encoder, consistently improving vision-language model performance across 15 benchmarks relative to a strong SigLIP baseline.

This paper aims to improve the performance of video multimodal large language models (MLLM) via long and rich context (LRC) modeling. As a result, we develop a new version of InternVideo2.5 with a focus on enhancing the original MLLMs' ability to perceive fine-grained details and capture long-form temporal structure in videos. Specifically, our approach incorporates dense vision task annotations into MLLMs using direct preference optimization and develops compact spatiotemporal representations through adaptive hierarchical token compression. Experimental results demonstrate this unique design of LRC greatly improves the results of video MLLM in mainstream video understanding benchmarks (short & long), enabling the MLLM to memorize significantly longer video inputs (at least 6x longer than the original), and master specialized vision capabilities like object tracking and segmentation. Our work highlights the importance of multimodal context richness (length and fineness) in empowering MLLM's innate abilites (focus and memory), providing new insights for future research on video MLLM. Code and models are available at https://github.com/OpenGVLab/InternVideo/tree/main/InternVideo2.5

Self-supervised visual foundation models produce powerful embeddings that achieve remarkable performance on a wide range of downstream tasks. However, unlike vision-language models such as CLIP, self-supervised visual features are not readily aligned with language, hindering their adoption in open-vocabulary tasks. Our method, named dino.txt, unlocks this new ability for DINOv2, a widely used self-supervised visual encoder. We build upon the LiT training strategy, which trains a text encoder to align with a frozen vision model but leads to unsatisfactory results on dense tasks. We propose several key ingredients to improve performance on both global and dense tasks, such as concatenating the [CLS] token with the patch average to train the alignment and curating data using both text and image modalities. With these, we successfully train a CLIP-like model with only a fraction of the computational cost compared to CLIP while achieving state-of-the-art results in zero-shot classification and open-vocabulary semantic segmentation.

Current vision systems typically assign fixed-length representations to images, regardless of the information content. This contrasts with human intelligence - and even large language models - which allocate varying representational capacities based on entropy, context and familiarity. Inspired by this, we propose an approach to learn variable-length token representations for 2D images. Our encoder-decoder architecture recursively processes 2D image tokens, distilling them into 1D latent tokens over multiple iterations of recurrent rollouts. Each iteration refines the 2D tokens, updates the existing 1D latent tokens, and adaptively increases representational capacity by adding new tokens. This enables compression of images into a variable number of tokens, ranging from 32 to 256. We validate our tokenizer using reconstruction loss and FID metrics, demonstrating that token count aligns with image entropy, familiarity and downstream task requirements. Recurrent token processing with increasing representational capacity in each iteration shows signs of token specialization, revealing potential for object / part discovery.

We propose a conditional positional encoding (CPE) scheme for vision Transformers. Unlike previous fixed or learnable positional encodings, which are pre-defined and independent of input tokens, CPE is dynamically generated and conditioned on the local neighborhood of the input tokens. As a result, CPE can easily generalize to the input sequences that are longer than what the model has ever seen during training. Besides, CPE can keep the desired translation-invariance in the image classification task, resulting in improved performance. We implement CPE with a simple Position Encoding Generator (PEG) to get seamlessly incorporated into the current Transformer framework. Built on PEG, we present Conditional Position encoding Vision Transformer (CPVT). We demonstrate that CPVT has visually similar attention maps compared to those with learned positional encodings and delivers outperforming results. Our code is available at https://github.com/Meituan-AutoML/CPVT .

Pioneering dual-encoder pre-training works (e.g., CLIP and ALIGN) have revealed the potential of aligning multi-modal representations with contrastive learning. However, these works require a tremendous amount of data and computational resources (e.g., billion-level web data and hundreds of GPUs), which prevent researchers with limited resources from reproduction and further exploration. To this end, we propose a stack of novel methods, which significantly cut down the heavy resource dependency and allow us to conduct dual-encoder multi-modal representation alignment with limited resources. Besides, we provide a reproducible baseline of competitive results, namely ZeroVL, with only 14M publicly accessible academic datasets and 8 V100 GPUs. Additionally, we collect 100M web data for pre-training, and achieve comparable or superior results than state-of-the-art methods, further proving the effectiveness of our methods on large-scale data. We hope that this work will provide useful data points and experience for future research in contrastive vision-language pre-training. Code is available at https://github.com/zerovl/ZeroVL.

Pre-trained vision encoders like DINOv2 have demonstrated exceptional performance on unimodal tasks. However, we observe that their feature representations are poorly aligned across different modalities. For instance, the feature embedding for an RGB image and its corresponding depth map of the same scene exhibit a cosine similarity that is nearly identical to that of two random, unrelated images. To address this, we propose the Omnivorous Vision Encoder, a novel framework that learns a modality-agnostic feature space. We train the encoder with a dual objective: first, to maximize the feature alignment between different modalities of the same scene; and second, a distillation objective that anchors the learned representations to the output of a fully frozen teacher such as DINOv2. The resulting student encoder becomes "omnivorous" by producing a consistent, powerful embedding for a given scene, regardless of the input modality (RGB, Depth, Segmentation, etc.). This approach enables robust cross-modal understanding while retaining the discriminative semantics of the original foundation model.

Multimodal retrieval still leans on embedding-based models like CLIP for fast vector search over pre-computed image embeddings. Yet, unlike text retrieval, where joint-encoder rerankers are standard, comparable vision--language rerankers are largely absent. We find that seminal joint encoders such as BLIP are severely bottlenecked by an expensive visual feature-extraction stage, preventing practical deployment at scale. Motivated by this bottleneck, we introduce EDJE, an Efficient Discriminative Joint Encoder that precomputes vision tokens offline and compresses them via a lightweight attention-based adapter, so online inference runs only a compact joint encoder over a small set of visual tokens plus the text. EDJE preserves strong retrieval performance while drastically reducing storage and online compute, enabling high-throughput inference. Specifically, EDJE processes 50k image--text pairs/second while requiring 49kB of disk storage per image, matching prior art on Flickr (zero-shot) and COCO (fine-tuned) retrieval. The implementation and checkpoints will be made publicly available shortly.

In-context prompting in large language models (LLMs) has become a prevalent approach to improve zero-shot capabilities, but this idea is less explored in the vision domain. Existing visual prompting methods focus on referring segmentation to segment the most relevant object, falling short of addressing many generic vision tasks like open-set segmentation and detection. In this paper, we introduce a universal visual in-context prompting framework for both tasks. In particular, we build on top of an encoder-decoder architecture, and develop a versatile prompt encoder to support a variety of prompts like strokes, boxes, and points. We further enhance it to take an arbitrary number of reference image segments as the context. Our extensive explorations show that the proposed visual in-context prompting elicits extraordinary referring and generic segmentation capabilities to refer and detect, yielding competitive performance to close-set in-domain datasets and showing promising results on many open-set segmentation datasets. By joint training on COCO and SA-1B, our model achieves 57.7 PQ on COCO and 23.2 PQ on ADE20K. Code will be available at https://github.com/UX-Decoder/DINOv.

3D dense captioning is a task involving the localization of objects and the generation of descriptions for each object in a 3D scene. Recent approaches have attempted to incorporate contextual information by modeling relationships with object pairs or aggregating the nearest neighbor features of an object. However, the contextual information constructed in these scenarios is limited in two aspects: first, objects have multiple positional relationships that exist across the entire global scene, not only near the object itself. Second, it faces with contradicting objectives--where localization and attribute descriptions are generated better with tight localization, while descriptions involving global positional relations are generated better with contextualized features of the global scene. To overcome this challenge, we introduce BiCA, a transformer encoder-decoder pipeline that engages in 3D dense captioning for each object with Bi-directional Contextual Attention. Leveraging parallelly decoded instance queries for objects and context queries for non-object contexts, BiCA generates object-aware contexts, where the contexts relevant to each object is summarized, and context-aware objects, where the objects relevant to the summarized object-aware contexts are aggregated. This extension relieves previous methods from the contradicting objectives, enhancing both localization performance and enabling the aggregation of contextual features throughout the global scene; thus improving caption generation performance simultaneously. Extensive experiments on two of the most widely-used 3D dense captioning datasets demonstrate that our proposed method achieves a significant improvement over prior methods.

Hypothesis. Artificial general intelligence is, at its core, a compression problem. Effective compression demands resonance: deep learning scales best when its architecture aligns with the fundamental structure of the data. These are the fundamental principles. Yet, modern vision architectures have strayed from these truths: visual signals are highly redundant, while discriminative information, the surprise, is sparse. Current models process dense pixel grids uniformly, wasting vast compute on static background rather than focusing on the predictive residuals that define motion and meaning. We argue that to solve visual understanding, we must align our architectures with the information-theoretic principles of video, i.e., Codecs. Method. OneVision-Encoder encodes video by compressing predictive visual structure into semantic meaning. By adopting Codec Patchification, OV-Encoder abandons uniform computation to focus exclusively on the 3.1%-25% of regions rich in signal entropy. To unify spatial and temporal reasoning under irregular token layouts, OneVision-Encoder employs a shared 3D RoPE and is trained with a large-scale cluster discrimination objective over more than one million semantic concepts, jointly capturing object permanence and motion dynamics. Evidence. The results validate our core hypothesis: efficiency and accuracy are not a trade-off; they are positively correlated. When integrated into LLM, it consistently outperforms strong vision backbones such as Qwen3-ViT and SigLIP2 across 16 image, video, and document understanding benchmarks, despite using substantially fewer visual tokens and pretraining data. Notably, on video understanding tasks, OV-Encoder achieves an average improvement of 4.1% over Qwen3-ViT. Codec-aligned, patch-level sparsity is a foundational principle, enabling OV-Encoder as a scalable engine for next-generation visual generalists.

We propose the Large View Synthesis Model (LVSM), a novel transformer-based approach for scalable and generalizable novel view synthesis from sparse-view inputs. We introduce two architectures: (1) an encoder-decoder LVSM, which encodes input image tokens into a fixed number of 1D latent tokens, functioning as a fully learned scene representation, and decodes novel-view images from them; and (2) a decoder-only LVSM, which directly maps input images to novel-view outputs, completely eliminating intermediate scene representations. Both models bypass the 3D inductive biases used in previous methods -- from 3D representations (e.g., NeRF, 3DGS) to network designs (e.g., epipolar projections, plane sweeps) -- addressing novel view synthesis with a fully data-driven approach. While the encoder-decoder model offers faster inference due to its independent latent representation, the decoder-only LVSM achieves superior quality, scalability, and zero-shot generalization, outperforming previous state-of-the-art methods by 1.5 to 3.5 dB PSNR. Comprehensive evaluations across multiple datasets demonstrate that both LVSM variants achieve state-of-the-art novel view synthesis quality. Notably, our models surpass all previous methods even with reduced computational resources (1-2 GPUs). Please see our website for more details: https://haian-jin.github.io/projects/LVSM/ .

High temporal resolution is essential for capturing fine-grained details in video understanding. However, current video large language models (VLLMs) and benchmarks mostly rely on low-frame-rate sampling, such as uniform sampling or keyframe selection, discarding dense temporal information. This compromise avoids the high cost of tokenizing every frame, which otherwise leads to redundant computation and linear token growth as video length increases. While this trade-off works for slowly changing content, it fails for tasks like lecture comprehension, where information appears in nearly every frame and requires precise temporal alignment. To address this gap, we introduce Dense Video Understanding (DVU), which enables high-FPS video comprehension by reducing both tokenization time and token overhead. Existing benchmarks are also limited, as their QA pairs focus on coarse content changes. We therefore propose DIVE (Dense Information Video Evaluation), the first benchmark designed for dense temporal reasoning. To make DVU practical, we present Gated Residual Tokenization (GRT), a two-stage framework: (1) Motion-Compensated Inter-Gated Tokenization uses pixel-level motion estimation to skip static regions during tokenization, achieving sub-linear growth in token count and compute. (2) Semantic-Scene Intra-Tokenization Merging fuses tokens across static regions within a scene, further reducing redundancy while preserving dynamic semantics. Experiments on DIVE show that GRT outperforms larger VLLM baselines and scales positively with FPS. These results highlight the importance of dense temporal information and demonstrate that GRT enables efficient, scalable high-FPS video understanding.

Contrastive learning methods for unsupervised visual representation learning have reached remarkable levels of transfer performance. We argue that the power of contrastive learning has yet to be fully unleashed, as current methods are trained only on instance-level pretext tasks, leading to representations that may be sub-optimal for downstream tasks requiring dense pixel predictions. In this paper, we introduce pixel-level pretext tasks for learning dense feature representations. The first task directly applies contrastive learning at the pixel level. We additionally propose a pixel-to-propagation consistency task that produces better results, even surpassing the state-of-the-art approaches by a large margin. Specifically, it achieves 60.2 AP, 41.4 / 40.5 mAP and 77.2 mIoU when transferred to Pascal VOC object detection (C4), COCO object detection (FPN / C4) and Cityscapes semantic segmentation using a ResNet-50 backbone network, which are 2.6 AP, 0.8 / 1.0 mAP and 1.0 mIoU better than the previous best methods built on instance-level contrastive learning. Moreover, the pixel-level pretext tasks are found to be effective for pre-training not only regular backbone networks but also head networks used for dense downstream tasks, and are complementary to instance-level contrastive methods. These results demonstrate the strong potential of defining pretext tasks at the pixel level, and suggest a new path forward in unsupervised visual representation learning. Code is available at https://github.com/zdaxie/PixPro.

This paper investigates two techniques for developing efficient self-supervised vision transformers (EsViT) for visual representation learning. First, we show through a comprehensive empirical study that multi-stage architectures with sparse self-attentions can significantly reduce modeling complexity but with a cost of losing the ability to capture fine-grained correspondences between image regions. Second, we propose a new pre-training task of region matching which allows the model to capture fine-grained region dependencies and as a result significantly improves the quality of the learned vision representations. Our results show that combining the two techniques, EsViT achieves 81.3% top-1 on the ImageNet linear probe evaluation, outperforming prior arts with around an order magnitude of higher throughput. When transferring to downstream linear classification tasks, EsViT outperforms its supervised counterpart on 17 out of 18 datasets. The code and models are publicly available: https://github.com/microsoft/esvit

We introduce a novel sequential modeling approach which enables learning a Large Vision Model (LVM) without making use of any linguistic data. To do this, we define a common format, "visual sentences", in which we can represent raw images and videos as well as annotated data sources such as semantic segmentations and depth reconstructions without needing any meta-knowledge beyond the pixels. Once this wide variety of visual data (comprising 420 billion tokens) is represented as sequences, the model can be trained to minimize a cross-entropy loss for next token prediction. By training across various scales of model architecture and data diversity, we provide empirical evidence that our models scale effectively. Many different vision tasks can be solved by designing suitable visual prompts at test time.

Despite the remarkable success of the LLaVA architecture for vision-language tasks, its design inherently struggles to effectively integrate visual features due to the inherent mismatch between text and vision modalities. We tackle this issue from a novel perspective in which the LLM not only serves as a language model but also a powerful vision encoder. To this end, we present LLaViT - Large Language Models as extended Vision Transformers - which enables the LLM to simultaneously function as a vision encoder through three key modifications: (1) learning separate QKV projections for vision modality, (2) enabling bidirectional attention on visual tokens, and (3) incorporating both global and local visual representations. Through extensive controlled experiments on a wide range of LLMs, we demonstrate that LLaViT significantly outperforms the baseline LLaVA method on a multitude of benchmarks, even surpassing models with double its parameter count, establishing a more effective approach to vision-language modeling.

Efficient and accurate feed-forward multi-view reconstruction has long been an important task in computer vision. Recent transformer-based models like VGGT and pi^3 have achieved impressive results with simple architectures, yet they face an inherent runtime bottleneck, due to the quadratic complexity of the global attention layers, that limits the scalability to large image sets. In this paper, we empirically analyze the global attention matrix of these models and observe that probability mass concentrates on a small subset of patch-patch interactions that correspond to cross-view geometric matches. Motivated by the structured attention and inspired by recent advancement in large language models, we propose a replacement for the dense global attention operation based on highly optimized block-sparse kernels, yielding up to 4times faster inference with comparable task performance. Our retrofit requires no retraining of the backbone, extends to both VGGT and pi^3, and supports large image collections. Evaluations on a comprehensive suite of multi-view benchmarks demonstrate the effectiveness of our approach.

Convolutional networks are powerful visual models that yield hierarchies of features. We show that convolutional networks by themselves, trained end-to-end, pixels-to-pixels, exceed the state-of-the-art in semantic segmentation. Our key insight is to build "fully convolutional" networks that take input of arbitrary size and produce correspondingly-sized output with efficient inference and learning. We define and detail the space of fully convolutional networks, explain their application to spatially dense prediction tasks, and draw connections to prior models. We adapt contemporary classification networks (AlexNet, the VGG net, and GoogLeNet) into fully convolutional networks and transfer their learned representations by fine-tuning to the segmentation task. We then define a novel architecture that combines semantic information from a deep, coarse layer with appearance information from a shallow, fine layer to produce accurate and detailed segmentations. Our fully convolutional network achieves state-of-the-art segmentation of PASCAL VOC (20% relative improvement to 62.2% mean IU on 2012), NYUDv2, and SIFT Flow, while inference takes one third of a second for a typical image.

Recently, Vision Large Language Models (VLLMs) integrated with vision encoders have shown promising performance in vision understanding. The key of VLLMs is to encode visual content into sequences of visual tokens, enabling VLLMs to simultaneously process both visual and textual content. However, understanding videos, especially long videos, remain a challenge to VLLMs as the number of visual tokens grows rapidly when encoding videos, resulting in the risk of exceeding the context window of VLLMs and introducing heavy computation burden. To restrict the number of visual tokens, existing VLLMs either: (1) uniformly downsample videos into a fixed number of frames or (2) reducing the number of visual tokens encoded from each frame. We argue the former solution neglects the rich temporal cue in videos and the later overlooks the spatial details in each frame. In this work, we present Balanced-VLLM (B-VLLM): a novel VLLM framework that aims to effectively leverage task relevant spatio-temporal cues while restricting the number of visual tokens under the VLLM context window length. At the core of our method, we devise a text-conditioned adaptive frame selection module to identify frames relevant to the visual understanding task. The selected frames are then de-duplicated using a temporal frame token merging technique. The visual tokens of the selected frames are processed through a spatial token sampling module and an optional spatial token merging strategy to achieve precise control over the token count. Experimental results show that B-VLLM is effective in balancing the number of frames and visual tokens in video understanding, yielding superior performance on various video understanding benchmarks. Our code is available at https://github.com/zhuqiangLu/B-VLLM.

This paper investigates the capability of plain Vision Transformers (ViTs) for semantic segmentation using the encoder-decoder framework and introduces SegViTv2. In this study, we introduce a novel Attention-to-Mask (\atm) module to design a lightweight decoder effective for plain ViT. The proposed ATM converts the global attention map into semantic masks for high-quality segmentation results. Our decoder outperforms the popular decoder UPerNet using various ViT backbones while consuming only about 5% of the computational cost. For the encoder, we address the concern of the relatively high computational cost in the ViT-based encoders and propose a Shrunk++ structure that incorporates edge-aware query-based down-sampling (EQD) and query-based upsampling (QU) modules. The Shrunk++ structure reduces the computational cost of the encoder by up to 50% while maintaining competitive performance. Furthermore, we propose to adapt SegViT for continual semantic segmentation, demonstrating nearly zero forgetting of previously learned knowledge. Experiments show that our proposed SegViTv2 surpasses recent segmentation methods on three popular benchmarks including ADE20k, COCO-Stuff-10k and PASCAL-Context datasets. The code is available through the following link: https://github.com/zbwxp/SegVit.

Customization of text-to-image models enables users to insert custom concepts and generate the concepts in unseen settings. Existing methods either rely on costly test-time optimization or train encoders on single-image training datasets without multi-image supervision, leading to worse image quality. We propose a simple approach that addresses both limitations. We first leverage existing text-to-image models and 3D datasets to create a high-quality Synthetic Customization Dataset (SynCD) consisting of multiple images of the same object in different lighting, backgrounds, and poses. We then propose a new encoder architecture based on shared attention mechanisms that better incorporate fine-grained visual details from input images. Finally, we propose a new inference technique that mitigates overexposure issues during inference by normalizing the text and image guidance vectors. Through extensive experiments, we show that our model, trained on the synthetic dataset with the proposed encoder and inference algorithm, outperforms existing tuning-free methods on standard customization benchmarks.

As the key component in multimodal large language models (MLLMs), the ability of the visual encoder greatly affects MLLM's understanding on diverse image content. Although some large-scale pretrained vision encoders such as vision encoders in CLIP and DINOv2 have brought promising performance, we found that there is still no single vision encoder that can dominate various image content understanding, e.g., the CLIP vision encoder leads to outstanding results on general image understanding but poor performance on document or chart content. To alleviate the bias of CLIP vision encoder, we first delve into the inherent behavior of different pre-trained vision encoders and then propose the MoVA, a powerful and novel MLLM, adaptively routing and fusing task-specific vision experts with a coarse-to-fine mechanism. In the coarse-grained stage, we design a context-aware expert routing strategy to dynamically select the most suitable vision experts according to the user instruction, input image, and expertise of vision experts. This benefits from the powerful model function understanding ability of the large language model (LLM) equipped with expert-routing low-rank adaptation (LoRA). In the fine-grained stage, we elaborately conduct the mixture-of-vision-expert adapter (MoV-Adapter) to extract and fuse task-specific knowledge from various experts. This coarse-to-fine paradigm effectively leverages representations from experts based on multimodal context and model expertise, further enhancing the generalization ability. We conduct extensive experiments to evaluate the effectiveness of the proposed approach. Without any bells and whistles, MoVA can achieve significant performance gains over current state-of-the-art methods in a wide range of challenging multimodal benchmarks. Codes and models will be available at https://github.com/TempleX98/MoVA.

Deploying high-performance dense prediction models on resource-constrained edge devices remains challenging due to strict limits on computation and memory. In practice, lightweight systems for object detection, instance segmentation, and pose estimation are still dominated by CNN-based architectures such as YOLO, while compact Vision Transformers (ViTs) often struggle to achieve similarly strong accuracy efficiency tradeoff, even with large scale pretraining. We argue that this gap is largely due to insufficient task specific representation learning in small scale ViTs, rather than an inherent mismatch between ViTs and edge dense prediction. To address this issue, we introduce EdgeCrafter, a unified compact ViT framework for edge dense prediction centered on ECDet, a detection model built from a distilled compact backbone and an edge-friendly encoder decoder design. On the COCO dataset, ECDet-S achieves 51.7 AP with fewer than 10M parameters using only COCO annotations. For instance segmentation, ECInsSeg achieves performance comparable to RF-DETR while using substantially fewer parameters. For pose estimation, ECPose-X reaches 74.8 AP, significantly outperforming YOLO26Pose-X (71.6 AP) despite the latter's reliance on extensive Objects365 pretraining. These results show that compact ViTs, when paired with task-specialized distillation and edge-aware design, can be a practical and competitive option for edge dense prediction. Code is available at: https://intellindust-ai-lab.github.io/projects/EdgeCrafter/

Recent advances in Vision-Language Models (VLMs) have enabled unified understanding across text and images, yet equipping these models with robust image generation capabilities remains challenging. Existing approaches often rely on reconstruction-oriented autoencoders or complex bridging mechanisms, leading to misalignment between understanding and generation representations, or architectural complexity. In this work, we propose VUGEN, a novel framework that explicitly leverages VLM's pretrained visual understanding priors for efficient and high-quality image generation. Our approach first transforms the high-dimensional latent space of the VLM's native vision encoder into a lower-dimensional, tractable distribution that maximally preserves visual information. The VLM is then trained to sample within this reduced latent space, ensuring alignment with its visual understanding capabilities. Finally, a dedicated pixel decoder maps these generated latents back to the image space. We find that a VAE-free pixel diffusion decoder to be on par or better than commonly used complex latent diffusion decoders that internally rely on VAE latents. Extensive experiments demonstrate that VUGEN achieves superior image generation performance, improving DPG Bench from 71.17 to 74.32 and FID from 11.86 to 9.06 on COCO, while fully preserving the VLM's original understanding capabilities.

Reading dense text and locating objects within images are fundamental abilities for Large Vision-Language Models (LVLMs) tasked with advanced jobs. Previous LVLMs, including superior proprietary models like GPT-4o, have struggled to excel in both tasks simultaneously. Moreover, previous LVLMs with fine-grained perception cost thousands of tokens per image, making them resource-intensive. We present TextHawk2, a bilingual LVLM featuring efficient fine-grained perception and demonstrating cutting-edge performance across general-purpose, OCR, and grounding tasks with 16 times fewer image tokens. Critical improvements include: (1) Token Compression: Building on the efficient architecture of its predecessor, TextHawk2 significantly reduces the number of tokens per image by 16 times, facilitating training and deployment of the TextHawk series with minimal resources. (2) Visual Encoder Reinforcement: We enhance the visual encoder through LVLM co-training, unlocking its potential for previously unseen tasks like Chinese OCR and grounding. (3) Data Diversity: We maintain a comparable scale of 100 million samples while diversifying the sources of pre-training data. We assess TextHawk2 across multiple benchmarks, where it consistently delivers superior performance and outperforms closed-source models of similar scale, such as achieving 78.4% accuracy on OCRBench, 81.4% accuracy on ChartQA, 89.6% ANLS on DocVQA, and 88.1% accuracy@0.5 on RefCOCOg-test.

Long-term video understanding requires interpreting complex temporal events and reasoning over procedural activities. While instructional video corpora, like HowTo100M, offer rich resources for model training, they present significant challenges, including noisy ASR transcripts and inconsistent temporal alignments between narration and visual content. In this work, we introduce an automated, training-free pipeline to extract high-quality procedural annotations from in-the-wild instructional videos. Our approach segments videos into coherent shots, filters poorly aligned content, and leverages state-of-the-art multimodal and large language models (Qwen2.5-VL and DeepSeek-R1) to generate structured, temporally grounded procedural steps. This pipeline yields DenseStep2M, a large-scale dataset comprising approximately 100K videos and 2M detailed instructional steps, designed to support comprehensive long-form video understanding. To rigorously evaluate our pipeline, we curate DenseCaption100, a benchmark of high-quality, human-written captions. Evaluations demonstrate strong alignment between our auto-generated steps and human annotations. Furthermore, we validate the utility of DenseStep2M across three core downstream tasks: dense video captioning, procedural step grounding, and cross-modal retrieval. Models fine-tuned on DenseStep2M achieve substantial gains in captioning quality and temporal localization, while exhibiting robust zero-shot generalization across egocentric, exocentric, and mixed-perspective domains. These results underscore the effectiveness of DenseStep2M in facilitating advanced multimodal alignment and long-term activity reasoning. Our dataset is available at https://huggingface.co/datasets/mingjige/DenseStep2M.

In this paper, we propose a post-training quantization framework of large vision-language models (LVLMs) for efficient multi-modal inference. Conventional quantization methods sequentially search the layer-wise rounding functions by minimizing activation discretization errors, which fails to acquire optimal quantization strategy without considering cross-layer dependency. On the contrary, we mine the cross-layer dependency that significantly influences discretization errors of the entire vision-language model, and embed this dependency into optimal quantization strategy searching with low search cost. Specifically, we observe the strong correlation between the activation entropy and the cross-layer dependency concerning output discretization errors. Therefore, we employ the entropy as the proxy to partition blocks optimally, which aims to achieve satisfying trade-offs between discretization errors and the search cost. Moreover, we optimize the visual encoder to disentangle the cross-layer dependency for fine-grained decomposition of search space, so that the search cost is further reduced without harming the quantization accuracy. Experimental results demonstrate that our method compresses the memory by 2.78x and increase generate speed by 1.44x about 13B LLaVA model without performance degradation on diverse multi-modal reasoning tasks. Code is available at https://github.com/ChangyuanWang17/QVLM.