Vision-Language Pretraining

Synthetic hard negatives generated in the representation space have proven effective for unimodal self-supervised learning, but transferring this idea to vision-language pretraining is not straightforward. We analyze six representation-space synthesis strategies and identify two failure modes in their transfer to vision-language pretraining: cross-modal constructions that produce overly easy negatives or pull them toward the query, and intra-modal constructions that incorporate the matched positive. We also observe logit-scale saturation when training with synthetic hard negatives and a learnable temperature, and find that fixing the temperature improves downstream performance. Using this geometric analysis we propose SNAP, which generates intra-modal hard negatives that never involve the positive from either modality, avoiding both failure modes entirely. SNAP is model-agnostic, requires no external generative models, and adds less than 10% training time overhead. Evaluated on top of CLIP and FLIP across multiple architectures and datasets, SNAP delivers consistent improvements on zero-shot retrieval, zero-shot classification, and linear probe evaluation.

Vision-language pre-training has reshaped image clustering, giving rise to language-assisted image clustering (LaIC), which leverages textual semantics to complement visual representations. Despite the rapid proliferation of LaIC methods, it remains unclear how much LaIC has actually advanced image clustering, as existing studies generally suffer from major limitations, including inconsistent experimental settings, inadequate dataset selection, and limited evaluation dimensions. To address this gap, we introduce VLM4Cluster, a comprehensive benchmark for image clustering in the era of pre-trained vision-language models (VLMs). VLM4Cluster implements 17 representative methods spanning classical, deep, and language-assisted image clustering, and evaluates them on 20 datasets covering classical, challenging, fine-grained, large-scale, and out-of-distribution settings. Beyond effectiveness, VLM4Cluster systematically investigates image clustering along three complementary dimensions: robustness to adversarial perturbations, generalization under distribution shifts, and computational efficiency. Our study shows that LaIC substantially advances the clustering performance frontier on many semantically demanding benchmarks, generally exhibits stronger generalization under distribution shifts, and achieves a more favorable effectiveness-efficiency trade-off. However, its gains become less consistent on large-scale and fine-grained datasets, while language assistance does not systematically reduce sensitivity to adversarial perturbations. VLM4Cluster is released at https://github.com/YuanweiHuu/VLM4Cluster.

Vision-language (VL) pretraining using paired chest X-ray (CXR) images and radiology reports has shown strong potential for medical image understanding. However, existing methods often remain dependent on task-specific finetuning because radiology reports are lengthy, clinically dense, and difficult to align with simple zero-shot prompts. Recent sentence-level approaches partially address this limitation using clinical phrases extracted by large language models (LLMs), but they largely overlook the intrinsic characteristics of radiology discourse. In particular, limited positive-pair diversity constrains further gains, while clinically equivalent sentences frequently recur across patients, creating false negatives in contrastive learning. To address these issues, we propose SentZero, an enhanced sentence-centric VL pretraining framework for zero-shot, multi-task CXR analysis. SentZero introduces LLM-based abstract-level sentence structuring and mapping to expand positive-pair diversity, together with an additional loss term to mitigate false negatives. We further introduce sentence-conditioned residual modulation of visual embeddings, enabling visual features to adapt to the semantic characteristics of each input sentence. Across diverse downstream tasks and datasets, SentZero improves zero-shot generalization and outperforms prior multi-task zero-shot methods.

Medical vision-language pre-training largely follows a visit-level image-report matching paradigm, aligning paired images and reports at individual visits. While effective for static cross-modal correspondence, this paradigm provides limited supervision for longitudinal clinical change, such as whether abnormalities improve, remain stable, or worsen over time. Learning such change is challenging because temporal semantics are implicit in free-text reports, and different abnormalities within the same patient may evolve asynchronously or even in opposite directions. We propose MedCTA, which reframes medical vision-language pre-training from visit-level cross-modal matching to learning clinical change. Rather than compressing a patient history into a single temporal representation, MedCTA models clinical change at two complementary scopes. At the abnormality scope, clinically grounded queries construct abnormality-conditioned visual and textual trajectories to capture heterogeneous abnormality evolution. At the patient-course scope, global image and report sequences are modeled to capture overall clinical progression beyond any individual abnormality. Structured trend supervision is extracted from longitudinal reports by an offline LLM parser, removing the need for manual temporal annotations. Combined with static image-report alignment, MedCTA learns representations that preserve visit-level cross-modal correspondence while encoding longitudinal change semantics. Experiments on temporal image classification, image-text retrieval, and zero-shot classification show consistent gains over strong medical vision-language baselines.

Long-tailed chest X-ray classification requires visual representations that capture both common abnormalities and subtle, infrequent findings. We propose Med-AR-8B and Med-AR-2B, two radiology-native autoregressive vision-language models pretrained with structured reports, abnormality-focused text, and region annotations. We evaluate the transfer of their visual encoders to multi-label classification against contrastive, self-supervised, and supervised pretrained encoders, including Med-CLIP, CheXFound, EVA-Base, ARK, and BioViL-T, using a common ML-Decoder classification head. To assess fine-grained recognition, we also construct LLM-expanded, report-derived label sets for MIMIC-CXR and CheXpert. Across PadChest, MIMIC-CXR, and CheXpert, Med-AR-8B outperforms Med-CLIP in mean AUROC and AUPRC for head, medium, and tail findings. On MIMIC-CXR, it increases tail-label mean AUPRC from 0.1033 to 0.1441. Med-AR-2B achieves the strongest discrimination results on PadChest. Across the broader encoder comparison, a Med-AR variant achieves the highest mean AUROC and AUPRC in every reported prevalence group on each public dataset. Both Med-AR variants also achieve lower excess area under the risk-coverage curve than Med-CLIP on all three public datasets, indicating improved selective-prediction performance under the evaluated protocol. Internal results are metric-dependent, with Med-CLIP retaining advantages in overall and tail AUPRC and in selective prediction. These findings establish Med-AR as a strong pretraining recipe for long-tailed chest X-ray classification on the evaluated public benchmarks and demonstrate the value of assessing discrimination and selective prediction together.

Structured visual reasoning, such as image puzzles, demands fine-grained visual perception, an ability current Vision Language Models (VLMs) lack. VLMs built on CLIP-based ViT backbones trade fine-grained detail for high-level semantics, and we show this loss propagates downstream. To recover it, we fuse DINOv3 and CleanDIFT representations into a perception encoder (DiffusedDINO) and align it with RoBERTa-L, yielding a text-aligned model TDDN that preserves this perceptual advantage: with frozen backbones and only $\sim$590K alignment pairs, TDDN matches CLIP on image-text retrieval, surpassing it on three of four settings. It does so while more than tripling CLIP's dense-prediction accuracy (ADE20K 5.20 $\to$ 18.11 mIoU, COCO-Stuff 7.35 $\to$ 24.44), despite CLIP's massive training corpus. TDDN leads on segmentation benchmarks among general-purpose contrastive encoders, including SigLIP$\,$2. We further introduce Puzzle Perception, a segmentation and visual question answering dataset that probes fine-grained spatial understanding, on which TDDN doubles CLIP's segmentation accuracy (11.04 $\to$ 22.51 mIoU).

Text-rich visual inputs require models that can read, retrieve, and compress language directly in pixel space, yet existing pixel-text encoders struggle with fixed resolution pretraining, visual shortcut learning, weak visual grounding, and multilingual visual text understanding. In this work, we investigate the fundamental design principles required for robust visual text representation learning. Through systematic controlled ablations, we identify four critical components: variable image resolutions and rendered font sizes provide spatial proxies for high-resolution document generalization; natural image-text pairs are indispensable for grounding and prevent text-only collapse; layout-aware rendering helps prevent pixel-level shortcuts; and a two-stage multilingual curriculum enables effective cross-lingual alignment. By integrating these principles into a scalable training recipe, we train Pixel Linguist II, a native-resolution vision encoder trained with on-the-fly rendering, unified contrastive grounding, and a multilingual curriculum over 280M training examples. Pixel Linguist II sets new state-of-the-art results on English, cross-lingual, and multilingual Visual STS and ViDoRe, while also enabling better MLLM downstream evaluation. Notably, Pixel Linguist II remains robust under 80% visual token compression, showing great promise for optical context compression. Our code and resources are available at https://github.com/Pixel-Linguist/Pixel-Linguist-II.

4D understanding and reasoning is a fundamental capability for embodied AI agents operating in dynamic physical environments. However, existing vision encoders are largely limited to static 2D images or 3D point clouds without temporal modeling, or to 2D videos that lack accurate geometric depth reasoning. Consequently, current approaches fail to jointly capture spatial structure and motion evolution in dynamic scenes. We present CL4D, the first foundational 4D vision encoder that directly operates on dynamic point clouds, trained with a contrastive learning objective to align spatio-temporal geometric representations with natural language descriptions. By learning a shared embedding space between text and 4D scene dynamics, CL4D enables zero-shot motion-to-text and text-to-motion retrieval in dynamic environments and serves as a foundational 4D vision encoder for downstream 4D vision-language tasks. Building on this encoder, we introduce 4DVLM, a 4D vision-language model that conditions language generation on dynamic geometric representations. 4DVLM is the first VLM designed to operate directly on 4D point clouds without relying on 2D images, 2D videos, or static 3D point clouds. We train CL4D and subsequently 4DVLM on a newly constructed dataset termed DynAction4D capturing diverse human motions across varying object interactions and scene environments. Extensive experiments across multiple 4D human action benchmarks demonstrate that CL4D achieves state-of-the-art performance, with improvements of approximately ~16.75% over prior methods. Furthermore, 4DVLM outperforms frontier video VLMs such as Gemini and GPT-5 even when these models are provided with RGB video sequences corresponding to the same scenes represented as 4D point clouds for 4DVLM.

Vision-language models remain underused in colonoscopy despite the rich expert descriptions recorded in routine reports. These reports document lesion appearance, size and location but summarise entire procedures rather than caption individual frames, leaving clinical findings only weakly linked to the corresponding images. Here we develop EndoCLIP, a colonoscopy vision-language foundation model trained on 125,756 lesion-level image-text pairs progressively recovered from 280,476 routine colonoscopy records. Across lesion-level image-text retrieval, structured report generation and six multi-centre clinical classification tasks, EndoCLIP outperforms general-purpose and biomedical vision-language encoders in both zero-shot and linear-probe settings. On benign-versus-malignant classification, its linear probe approaches the performance of expert readers in a blinded study involving 12 endoscopists. These results suggest that recovering finding-to-frame correspondence can transform routine documentation into scalable supervision, enabling clinical targets to be specified in language rather than separately annotated for each task.

While data curation for Vision Language Models (VLMs) is increasingly active, public practice for constructing pretraining mixtures remains largely heuristic: practitioners stack datasets that pass quality filters, set cross-domain ratios by intuition, and lack a principled, attributable criterion for admitting new data, while frontier recipes remain undisclosed. We formulate data construction as a systematic mixture-optimization problem and turn it into a reproducible engineering discipline by decoupling the mixture into two orthogonal sub-problems: inter-class ratios across capabilities and intra-class ratios within a category. For inter-class allocation, we use a single-variable iterative search; for intra-class composition, we apply a multidimensional, dataset-level assessment scoring Quality and Difficulty, and formulate selection as a constrained convex optimization with a diversity objective. The DecoupleMix framework delivers two critical capabilities: guiding what data to collect next and rendering dataset validation a controlled, attributable experiment. Experiments show our approach consistently surpasses heuristic baselines. Moreover, optimal ratios discovered on small-scale proxies transfer seamlessly to larger scales without retuning. Using 80B additional multimodal continue-pretraining tokens, our VLM is competitive with strong open-source models trained with substantially larger multimodal budgets.

Large Vision-Language Models (LVLMs) remain bottlenecked by massive computational footprints, precluding their deployment on resource-constrained edge devices. While efforts to compress LVLMs focus heavily on vision token reduction or smaller language models, the vision encoder is largely overlooked, typically deployed as a monolithic, computationally heavy feature extractor. Moreover, there is no previous effort that designs a vision encoder for LVLMs directly optimized for on-device latency. In this paper, we present UltraViT, a vision encoder for LVLMs, explicitly designed and optimized for on-device performance. Specifically, by taking into account real on-device latencies, we systematically design a pyramidal architecture that strategically integrates and adapts heterogeneous spatial mixers at the macro-block level. Furthermore, to pre-train UltraViT, we propose a novel two-stage generative pre-training strategy: cultivating rich spatial features via dense distillation, followed by direct generative supervision from a capacity-mixed frozen LLM. Compared to standard contrastive and SSL, we show that our pre-training is much more effective for achieving high-level semantic grounding for UltraViT needed for the subsequent generative multimodal alignment of LVLM training. Extensive experiments demonstrate that our on-device latency-informed design combined with our tailored training strategy establishes a new state-of-the-art for efficient LVLM encoding, significantly outperforming existing encoder-centric baselines while operating on-device at nearly 1.7xthe speed.

Contrastive Language-Image Pretraining learns a shared representation space through large-scale contrastive learning. However, existing methods that enforce global consistency regularization overlook a key challenge: the inherent information asymmetry between images and text: captions typically describe only one specific aspect of an image, thus images with similar visual content can be paired with completely divergent textual content and semantic information. Consequently, global regularizers inadvertently impose constraints between visually similar images whose captions describe divergent aspects, introducing semantic distortion into the representation space. We propose AspectCLIP, a framework that reformulates consistency regularization to respect this one-to-many structure. AspectCLIP first partitions training samples into attribute clusters based on textual similarity to identify aspect-coherent groups, then applies full cyclic consistency within each cluster while restricting cross-cluster regularization to prototype-level comparisons. This aspect-guided regularization enforces strict geometric alignment only when images and texts describe a consistent facet, while allowing flexibility across divergent aspects. Extensive experiments on downstream tasks demonstrate that AspectCLIP consistently outperforms traditional methods and achieves a more structured representation space.

Mainstream visual encoders are pretrained on natural images and cannot be effectively applied to document images without document-oriented adaptation, as dense text and fine-grained character strokes demand character-level visual perception. We present MonkeyOCRv2, a visual-text pretrained model for document AI. First, we construct MonkeyDoc v2, to our knowledge the largest document-image pretraining corpus, comprising 113 million images spanning 17 languages. Second, we propose a pretraining strategy that jointly learns image-to-text generation and pixel-level document reconstruction: the former aligns visual representations with textual content, while the latter preserves character strokes and layout details. Extensive experiments are conducted on five representative document analysis tasks, including text recognition, formula recognition, text detection, document tampering detection, and overlapping text segmentation. Replacing the original encoders with MonkeyOCRv2 consistently improves performance across all five tasks. Finally, we validate its effectiveness as the vision encoder of multimodal large language models on the more challenging tasks of document parsing and document understanding. Kept frozen and paired with a lightweight language model, it yields a 0.7B document parsing model that sets a new open-source state-of-the-art on MDPBench, a recent benchmark spanning digital-born and photographed documents across 17 languages, surpassing the previous best 3B dots.mocr by 2.8% absolute with a vision encoder roughly 11$\times$ smaller. The frozen encoder also powers a document understanding model that outperforms counterparts built on CLIP, DINO, and SAM across eight benchmarks under identical training settings. These results suggest that document-oriented visual pretraining can serve as a foundation for document intelligence in its own right.

Open-vocabulary dense perception (OVDP) aims to localize objects unseen during training by leveraging textual knowledge. Despite the remarkable progress of recent CLIP-based approaches, we identify a critical limitation: synonym-induced grounding inconsistency, where semantically equivalent expressions yield disparate spatial attention patterns. This inconsistency undermines the robustness and performance of existing methods in real-world OVDP applications. To address this issue, we propose SynCLIP, a Synonym-Coherent Language-Image Pretraining framework that enhances synonym-robust grounding for OVDP. SynCLIP introduces a Semantic-consistent Spatial Attention alignment (SSA) module to enhance spatial attention consistency by minimizing discrepancies between attention maps of original and synonymous expressions. Furthermore, a Spatial Attention Refinement (SAR) module selectively strengthens the most semantically relevant spatial regions within aligned maps for more precise and stable grounding. To support synonym-coherent pretraining, we also construct a Synonym-Enriched Visual Corpus (SEViC), which augments each category with multiple synonyms and textual definitions. Extensive experiments on multiple benchmarks demonstrate that SynCLIP substantially improves grounding consistency under diverse linguistic variants and achieves state-of-the-art performance among CLIP-based OVDP methods. Code is available at https://github.com/Justlovesmile/SynCLIP.

Vision-language pretraining remains dominated by contrastive objectives, whereas vision-only self-supervised learning has largely adopted non-contrastive methods. At the same time, the role of vision-language encoders has shifted: they are increasingly deployed not as zero-shot classifiers but as the frozen visual backbone of vision-language models and dense prediction systems, which consume the full grid of patch tokens rather than a single pooled embedding. We introduce LeVLJEPA, the first fully non-contrastive end-to-end vision-language pretraining method. LeVLJEPA learns through cross-modal prediction with stop-gradient targets and per-modality distributional regularization, without negatives, temperature, momentum encoder, or teacher-student schedule, and trains stably at large scale. We find that the resulting encoder provides markedly stronger dense semantic features for downstream use: as a frozen vision-language-model backbone, LeVLJEPA is the strongest of the evaluated encoders across GQA, VQAv2, and POPE under two distinct language models, and outperforms contrastive baselines on semantic segmentation, while remaining on par on global readouts such as linear probing. These results establish non-contrastive pretraining as an effective means of producing dense semantic vision features.

Vision-language foundation models have shown strong potential in medical image analysis. Although foundation models for ultrasound imaging have recently emerged, the domain remains particularly challenging due to severe speckle noise, acquisition variability, and subtle anatomical boundaries, leading to high inter-observer variability. Existing CLIP-based models rely primarily on global image-text alignment, limiting their sensitivity to clinically decisive local structures. We propose SonoCLIP, the first million-scale region-controllable fetal ultrasound vision-language foundation model that integrates segmentation masks as mask-channel visual prompts within the vision encoder, enabling joint global-local contrastive representation learning. To support scalable region-text alignment, we introduce a sigmoid-based pairwise contrastive loss that improves stability under large-scale supervision. We further curate a 1.44M-image multimodal fetal ultrasound dataset spanning 24 standard planes for large-scale pretraining. Extensive cross-center evaluations demonstrate that SonoCLIP achieves superior zero-shot transfer performance under both global and mask-guided inference, establishing a controllable and clinically oriented foundation model for fetal ultrasound analysis. Our code and data are available at https://github.com/Harrison-one/SonoCLIP.

Building performant Vision-Language Models (VLMs) requires carefully curating large-scale training datasets, yet the community lacks systematic benchmarks for evaluating such curation strategies. We introduce DataComp for VLMs (DCVLM), a benchmark for controlled data-centric experiments to improve VLM training. As part of DCVLM, we collect 160 datasets spanning four data types -- image-caption pairs, multimodal interleaved documents, text-only, and instruction-tuning data -- into a corpus of 6T multimodal tokens. DCVLM allows participants to test curation strategies (filtering, mixing, formatting, sampling) across 1B-8B models and 6.25B-200B token budgets. Models are then evaluated on a carefully selected suite of up to 52 downstream benchmarks across 9 domains. We conduct extensive experiments on DCVLM and find that data mixing, not filtering, is key to a high-quality training dataset: instruction-heavy mixtures scale better than caption-heavy ones, with gains widening at larger scales. The resulting dataset, DCVLM-Baseline, enables training an 8B VLM to 63.6% accuracy on our 33-task core suite with 200B training tokens. Compared to FineVision, the state-of-the-art open VLM training dataset, this represents an improvement of +5.4pp. DCVLM and all accompanying artifacts will be made publicly available at https://www.datacomp.ai/dcvlm/.

CLIP and its variants are widely adopted visual backbones in multimodal systems, but their pretraining remains dominated by descriptive image-text alignment. As downstream applications increasingly demand visually grounded commonsense inference and compositional reasoning, it remains unclear whether CLIP-style encoders can support such reasoning without architectural changes. To address this, we present ReasonCLIP-58M, a continual pretraining framework that integrates large-scale reasoning supervision into CLIP-style models through our two-stage strategy, which progressively integrates reasoning signals while preserving descriptive alignment, followed by category-structured reasoning supervision. To support this framework, we construct two complementary datasets and a benchmark: ReasonLite-42M, with open-form, visually verifiable reasoning captions; ReasonPro-16M, with category-specific reasoning supervision; and RCLIP-Bench for diagnostic evaluation of visually grounded reasoning. We train a family of ReasonCLIP that improves visually grounded commonsense and compositional reasoning while also enhancing zero-shot retrieval performance. As a drop-in visual encoder for multimodal large language models such as LLaVA-NeXT, ReasonCLIP delivers consistent gains without additional inference cost, demonstrating that structured reasoning supervision enhances the expressive capacity of CLIP-style visual representations. All datasets, models, and training code are available at https://github.com/RISys-Lab/ReasonCLIP.

Modern VLMs and VLA systems commonly adopt off-the-shelf ViTs such as SigLIP2 as visual encoders, but diverse downstream requirements in latency, temporal modeling, and VLM integration often call for customized SOTA-level ViTs. Training such encoders remains beyond the reach of much of the community, as it requires massive image-text data, while standard softmax attention makes high-resolution or dynamic-resolution pretraining prohibitively costly and often forces low-resolution pretraining followed by post-hoc adaptation. TuringViT addresses these challenges with three key designs: Turing Linear Attention (TLA) for efficient sequence modeling, VISTA-Curation to construct supervision-rich image-video training data, and native dynamic-resolution pretraining that supports flexible inputs from the start and transfers seamlessly to downstream VLMs. As a result, TuringViT outperforms leading open-source ViT baselines with only 10% of the data, achieves stronger downstream VLM performance, and delivers substantially better latency scaling on high-resolution inputs. Our scaling-law analysis further shows that TuringViT continues to improve predictably with curated data scale, far from saturation. Its fast adaptation, hardware-friendly design, and efficient deployment have made it a unified visual foundation across XPeng's AI systems. More broadly, TuringViT provides a reproducible pipeline that dramatically lowers the cost for the community to train, customize, and deploy SOTA-level ViTs, moving toward making such Vision Transformers accessible to all.

The availability of large amounts of clean data is paramount to training neural networks. However, at large scales, manual oversight is impractical, resulting in sizeable datasets that can be very noisy. Attempts to mitigate this obstacle to producing performant vision-language models have so far involved heuristics, curated reference datasets, and using pre-trained models. Here we propose a novel, bootstrapped method in which a CLIP model is trained on an evolving, self-selected dataset. This evolving dataset constitutes a balance of filtered, highly probable clean samples as well as diverse samples from the entire distribution. Our proposed Self-Filtering method iterates between training the model and selecting a subsequently improved data mixture. Training on vision-language datasets filtered by the proposed approach improves downstream performance without the need for additional data or pre-trained models.

Contrastive Language-Image Pre-training models are widely reused across downstream interfaces, including feature extraction, retrieval, reranking, and selection. Existing CLIP backdoor, however, usually validate attacks on a small attack-native task, leaving unclear whether the same poisoned checkpoint remains exposed, weakens, or becomes not applicable when reused through other interfaces. We introduce DIFE, a Deployment-Interface Footprint Evaluation framework that audits backdoored CLIP checkpoints across deployment interfaces. DIFE makes various evaluations comparable by specifying each interface's component readout, trigger channel, target event, reference condition, and metric. DIFE also introduces effective-footprint diagnosis to identify the reusable CLIP component or component combination that carries exposure and explains where risk transfers. Auditing reproduced CLIP backdoors with DIFE reveals a structured landscape: native success is not a checkpoint-level risk certificate, exposure follows component footprints, text-side poisoning does not yield textual-encoder control, and some coupled attacks remain mechanism-bound. This audit reveals a import gapin existing CLIP backdoors: a textual encoder that itself becomes a reusable carrier of adversarial behavior. We therefore introduce BadTextTower to fill this gap. BadTextTower produces strong text-conditioned retrieval, reranking, and selection exposure while leaving visual-only reuse nearly clean.

Vision-Language-Action (VLA) models benefit from large-scale and diverse embodied data, yet scaling robot trajectory collection is costly and labor-intensive. Recent advances show that large-scale egocentric human videos provide complementary real-world supervision in pretraining. However, joint training on human and robot data remains challenging due to divergences in action spaces, embodiment structures, temporal dynamics, and supervision quality. We introduce ACE-EGO-0, a unified VLA pretraining framework jointly leveraging heterogeneous data sources. To extract large-scale pretraining supervision from egocentric human videos, we build a scalable egocentric video-to-action pipeline that converts raw human videos into robot-format pseudo-action trajectories. To make these labels comparable with robot demonstrations, ACE-EGO-0 uses a unified action representation based on camera-space actions, morphology conditioning, and time-aligned action chunking. To robustly leverage noisy pseudo-action supervision from egocentric human videos, we formulate a reliability-aware training objective with a human auxiliary loss that concentrates supervision on reliable signals. We instantiate ACE-EGO-0 on 4.53K hours of robot and simulation data, together with 1.48K hours of pseudo-action-labeled egocentric human data. Experiments show that incorporating large-scale human supervision under reliability-aware weighting consistently improves both unified joint pretraining and supervised fine-tuning. ACE-EGO-0 achieves state-of-the-art performance on RoboCasa GR1 TableTop and RoboTwin 2.0, while demonstrating strong transfer to real-world bimanual manipulation.

Contrastively trained vision-language models like CLIP, have made remarkable progress in learning joint image-text representations, but still face challenges in compositional understanding. They often exhibit a "bag-of-words" behavior--struggling to capture the object relations, attribute-object bindings, and word order dependencies. This limitation arises not only from the reliance on global, single-vector representations for optimization, but also from the insufficient exploitation and modeling of the rich compositional information inherently present in paired image text data. In this work, we propose MACCO (MAsked Compositional Concept MOdeling), a framework that masks compositional concepts in one modality and reconstructs them conditioned on the full contextual information from the other, enabling the model to capture and align cross-modal compositional structures more effectively. To facilitate this process, we introduce two auxiliary objectives that jointly align and regularize masked features both inter-modally and intra-modally. Extensive experiments on five compositional benchmarks, along with in-depth analyses, demonstrate that our approach not only significantly enhances compositionality in VLMs but also improves their ability to capture syntactic structure and linguistic information. Additionally, the improved compositionality also benefits text-to-image generation and multimodal large language model. Code is available at https://github.com/hiker-lw/MACCO.

Modern Vision-Language-Action (VLA) models must bridge pretrained vision-language reasoning and precise continuous robot control. Existing action tokenizers discretize actions primarily for reconstruction, producing codes that preserve motion geometry but provide only weak semantic supervision to the backbone. We therefore formulate action tokenization not as mere compression, but as semantic interface learning between multimodal reasoning and executable control. To this end, we introduce X-Tokenizer, a lightweight encoder-Semantic Residual Quantization (SRQ)-decoder architecture that provides a shared action interface across diverse robotic arm embodiments. Its key component, SRQ, imposes an asymmetric structure on residual vector quantization: the first level is trained with Masked Action Modeling (MAM) to form a discrete action language that captures coarse motion intent, while deeper levels remain reconstruction-oriented residuals that preserve fine-grained details. To further align action tokens with multimodal semantics, X-Tokenizer is pretrained with contrastive alignment to the representation space of a pretrained foundation model and with next-frame vision-language feature prediction. Pretrained on 2.4M trajectories (2.0B action frames), a single frozen X-Tokenizer plugs into a mixed discrete-continuous VLA as a representation-shaping supervision signal. X-Tokenizer achieves top real-world aggregate and strong RoboTwin 2.0 simulation results. Outperforming FAST in multimodal grounding (+13.5%) and long-horizon tasks (+8.25), it shows that action tokenizers serve as semantic interfaces for VLA pretraining beyond mere action compression.

WALL-WM is a World Action Model that shifts video-action learning from chunk-centric optimization to event-grounded Vision-Language-Action pretraining, using semantically coherent action events as the atomic unit of learning. Existing WAMs commonly initialize from multimodal or video foundation models and then optimize fixed-length action chunks conditioned directly on the current observation and instruction. Although convenient, this chunk-centric formulation creates a fundamental granularity mismatch. Language describes semantic goals and events, vision evolves through continuous scene dynamics, and actions operate at control-level timescales; forcing all three into the same fixed-length prediction window turns VLA training into short-horizon correlation fitting. WALL-WM addresses this mismatch by organizing both supervision and data around semantic events. Specifically, it pairs event-grounded VLA pretraining with a data ecosystem built from event-level captions and cluster-balanced sampling, enabling scalable learning over diverse behaviors, scenes, and task structures. From the same event-pretrained backbone, WALL-WM supports two complementary inference modes. The event mode consumes next-event descriptions and enables variable-length execution chunks, while the unified mode uses a VLM with Staircase Decoding to condition conventional fixed-length chunk inference while preserving a gradient-continuous VLA path. Together with Muon-optimizer-based large-scale pretraining infrastructure, WALL-WM provides a practical scale-up recipe for general-purpose WAMs. Experiments show that WALL-WM generalizes broadly across language, scenes, and tasks, achieving state-of-the-art performance in large-scale real-world generalization evaluation.

Vision-Language Models (VLMs) largely follow the text-only LLM trajectory, excelling on English benchmarks but sharply degrading on low-resource languages, where neither large-scale image-text corpora nor culturally grounded evaluations exist. We present a systematic study of building a language-specific VLM for Romanian, covering the full pipeline from data construction to architectural choices. We translate established English VLM training and evaluation corpora into Romanian, applying machine translation to textual annotations and to in-image text, preserving visual grounding while adapting the textual content. Using this data, we train and ablate a series of VLMs to isolate the contribution of (i) vision backbones of varying scale and pretraining, (ii) language backbones from multilingual to Romanian-adapted LLMs, and (iii) OCR-style image-text data. We further curate HoraVQA, a culturally native evaluation set grounded in Romanian everyday scenes. Romanian-adapted VLMs consistently outperform their same-sized counterparts and, across all evaluated benchmarks, even surpass models from the next larger size category.

Large-scale Vision-Language-Action (VLA) pretraining is increasingly adopted as the foundation for robot policies, yet the evidence for pretrained VLAs is almost invariably reported after task-specific fine-tuning. This leaves a foundational question unanswered: does VLA pretraining itself yield executable robot behavior, or does it merely furnish a better initialization for downstream policy learning? We present Wall-OSS-0.5, an open-source 4B VLA built upon a 3B VLM backbone augmented with action-generation components, designed so that pretrained robotic capability is directly measurable on physical hardware. The model is pretrained across more than 20 embodiments, processing over one million robot trajectories per epoch alongside a grounded multimodal corpus. We adopt a gradient-bridged co-training recipe in which three objectives play distinct and complementary roles: discrete action prediction routes strong VLM-native gradients into the backbone, multimodal prediction preserves grounded vision-language understanding, and continuous flow matching serves as the deployment-time action interface. Before task-specific fine-tuning, the pretrained checkpoint achieves non-trivial zero-shot real-robot behavior, completing several tasks, including a held-out deformable manipulation task, at high task progress on a 17-task suite. After fine-tuning, the same checkpoint serves as a stronger adaptation prior, reaching 60.5% average task progress on 15 real-robot tasks and outperforming π_0.5 by 17.5%. Multimodal evaluations further confirm that action training does not erode grounded vision-language competence: the model preserves broad vision-language ability while strengthening embodied grounding. Together, these results reposition VLA pretraining from an initialization strategy to a directly testable, already useful source of robot capability.

Current vision-language models (VLMs) typically stitch together separate image encoders and language decoders via multi-stage alignment, a modular framework that inevitably fragments pixel-level signals across frames and scatters early pixel-word interactions. In parallel, native VLMs, despite impressive performance on single images, remain largely unexplored in multi-image, video understanding, and spatial intelligence. Hence, we introduce NEO-ov, a native foundation model that learns cross-frame and pixel-word correspondence end-to-end, without any external encoders, auxiliary adapters, or post-hoc fusion. By eliminating module boundaries entirely, NEO-ov enables fine-grained and unified spatiotemporal modeling to emerge natively inside the model. Notably, NEO-ov largely narrows the gap to modular counterparts while excelling at fine-grained visual perception, validating that native "one-vision" architectures are not only feasible but competitive at scale. Beyond empirical performance, we unveil systematic architectural analyses and detailed training recipes to facilitate subsequent native multimodal modeling. Our code and models are publicly available at: https://github.com/EvolvingLMMs-Lab/NEO.

Embodied Vision-Language Models (VLMs) have demonstrated impressive performance and generalization in robotics, particularly within Vision-Language-Action frameworks. However, a significant gap remains between the high-level semantic focus of standard text-guided pre-training paradigms and the low-level spatial and physical knowledge critical for execution in embodied environments. In this paper, we introduce GEM, a Generative-supervised Embodied vision-language Model designed to bridge this divide. We propose integrating a depth map generation task directly into the VLM pre-training phase. By training this generative objective jointly with the main model, we observe substantial improvements in embodied intelligence, significantly enhancing both semantic understanding and physical operation capabilities. To support this paradigm, we curate and release GEM-4M, a comprehensive large-scale dataset featuring a mixture of grounding, reasoning, and planning data paired with high-quality depth supervision. Extensive experiments demonstrate that GEM achieves state-of-the-art results across diverse embodied benchmarks. Furthermore, our deployed action model, GEM-VLA, exhibits vastly superior task execution abilities in both simulation environments and real-world evaluations. Code, models, and datasets are available at https://zhaorw02.github.io/GEM/

Vision-Language-Action (VLA) models widely adopt pretrained Vision-Language Models (VLMs) as policy backbones, yet it remains unclear what kind of pretrained VLM representation is useful as a VLA initialization. In this paper, we study VLA initialization as a controlled representation-design problem along three axes: capability-level embodied VQA supervision, parameter-update strategy, and robot-data pretraining. Our experiments show that the original pretrained VLM representation is a key source of action performance. However, embodied VQA adaptation does not yield uniform gains: its benefit depends on downstream bottlenecks, and gains from different capability domains are not simply additive. For update strategy, LoRA provides a more reliable initialization than Full Finetune, indicating that overly reshaping the pretrained representation can weaken VLA initialization. Robot-data pretraining further improves VLA initialization, with the strongest variant obtained by staged LoRA-based training. Together, these findings suggest that effective VLM-to-VLA adaptation should inject action-relevant embodied and robot-trajectory signals while preserving the pretrained VLM representation that remains useful for action learning.