arXiv subjects

Ziang Yan

Publications and source records attributed to Ziang Yan.

At least 19 recordsLinked to original sources

arXiv2026

Intern-S2-Preview: Scientific Agentic Foundation Model

Scientific discovery increasingly requires AI systems that can reason over scientific evidence of heterogeneous modalities, interact with scientific tools and environments, and sustain progress across long task horizons. We present Intern-S2-Preview, a series of scientific agentic foundation models designed to support multimodal scientific understanding, reasoning, generation, and long-horizon tasks. The training pipeline begins with scientific multimodal pre-training over rendered scientific documents, interleaved image-text data, and diverse scientific corpora. Starting from the pretrained checkpoint, we apply a unified post-training pipeline consisting of supervised fine-tuning, scalable multi-task reinforcement learning (RL), black- and white-box agentic RL, and on-policy distillation. This pipeline is supported by practical techniques that improve rollout and training stability and efficiency, including partial rollout with off-policy correction, adaptive length regularization, online speculative decoding, robust multi-task optimization, and trace-aware experience assembly for agentic tasks. At the architecture level, Intern-S2-Preview-397B extends time series modelling from efficient long-sequence understanding to numerical forecasting, while Memory Decoder is studied as a separate memory-augmented path for rapid scientific specialization without modifying the frozen 397B backbone. Evaluations across scientific, multimodal, agentic, and general-purpose benchmarks show that Intern-S2-Preview-397B achieves competitive or leading results in multiple settings. The time series modules improve scientific signal understanding and forecasting on SciTS, while the separate Intern-MemDec-4B extension improves the Biology-Instructions average score from 56.92 to 60.32 without modifying the frozen 397B backbone.

cs.LG↗

arXiv2026

TimeLens2: Generalist Video Temporal Grounding with Multimodal LLMs

Video multimodal large language models (MLLMs) can describe what happens in a video, but rarely identify when the supporting evidence occurs. We study generalist video temporal grounding, in which one model predicts a variable-cardinality set of evidence intervals across video lengths, domains, query forms, and viewpoints. Existing training strategies are misaligned with this set-valued task: long-video labels often rely on brittle one-pass annotation, while reinforcement-learning rewards either fail to distinguish non-overlapping predictions or require fragile segment matching. TimeLens2 treats temporal evidence as an interval set throughout supervision and optimization. TimeLens2-93K constructs reliable multi-span supervision through caption-derived proposals, independent localization, cross-agent consensus, semantic verification, and boundary refinement. Our temporal Wasserstein reward computes exact one-dimensional $W_1$ between uniform distributions over merged interval supports, providing dense, matching-free feedback under unequal cardinalities and equivalent fragmentation; temporal IoU complements it with precise-overlap feedback. Across seven benchmarks, TimeLens2-2B outperforms all size-matched baselines on every benchmark, while the 4B and 8B variants achieve state-of-the-art performance, surpassing open-source models with up to 397B parameters. The 2B, 4B, and 8B variants improve over their Qwen3-VL backbones by 14.2, 13.0, and 18.1 mIoU points, respectively.

cs.CV↗

arXiv2026

VideoChat3: Fully Open Video MLLM for Efficient and Generalist Video Understanding

Recent advances in video understanding have spanned motion, long video, and streaming interaction, driving this field toward real-world applications. Despite this progress, current open-source models remain limited in several ways. They often struggle to generalize across diverse video types, making them effective only in specific domains. High computational demands further restrict their efficiency and scalability. Moreover, most models are only partially open, with key components such as training code, strategy, or datasets unavailable, which hinders reproducibility and slows community-driven development. To address these issues, we introduce VideoChat3, a fully open, efficient, and generalist video-centric MLLM. VideoChat3 advances video understanding through two complementary designs. For efficiency, we introduce Inflated 3D Vision Transformer (I3D-ViT) and Adaptive Frame Resolution for Streaming Video Perception, which enables efficient spatiotemporal representation and reduces the cost of processing video inputs during training and inference. For effectiveness, we develop a scalable video data synthesis pipeline that curates three diverse, high-quality training datasets: VideoChat3-Academic2M, VideoChat3-LV116K, and VideoChat3-OL617K, covering general, long-form, and streaming video scenarios, improving the model's generalization across domains. By integrating these designs, VideoChat3 achieves a rare balance of broad generalization and computational efficiency. Experiments across general, long-form, and streaming benchmarks demonstrate that VideoChat3 surpasses prior open-source models with equal or larger parameter counts with only 4B parameters and higher efficiency.

cs.CV↗

arXiv2026

KiDS-Legacy: The consistency test of the large-scale structure with Bernardeau-Nishimichi-Taruya transform

We perform the first $k$-cut cosmic shear analysis of the KiDS-Legacy survey. This method uses the Bernardeau-Nishimichi-Taruya (BNT) transform to construct weak-lensing kernels that are more localised than conventional ones, and remove information from selected physical scales while retaining the constraining power of the targeted range. Removing the scale of $k \geq 0.33~\mathrm{Mpc}^{-1}$ from the KiDS-Legacy pseudo-$C_\ell$ data vector, and using a covariance matrix whose Gaussian component is computed from the theoretical data vector, we find $S_8 = 0.798 \pm 0.045$. This agrees with both the fiducial KiDS-Legacy bandpower result and our no-$k$-cut pseudo-$C_\ell$ posterior to within $0.1\sigma$, indicating no significant bias from nonlinear astrophysical feedback at the precision of KiDS-Legacy. We also study the case in which the Gaussian covariance is computed from the observed data vector. In this setup, the same scale cut of $k < 0.33~\mathrm{Mpc}^{-1}$ gives a much lower $S_8=0.717_{-0.046}^{+0.047}$. Further $k$-cut tests reveal a mild scale-dependent trend, with larger physical scales preferring lower $S_8$ values and a maximum low- versus high-$k$ deviation of $1.80\sigma$. Mock tests show that this behaviour is not produced by the covariance prescription or data vector alone, but may arise from their interplay. These results show that BNT $k$-cuts provide both a mitigation strategy for nonlinear systematics and a diagnostic of weak-lensing inference pipelines.

astro-ph.CO↗

arXiv2026

InternVideo3: Agentify Foundation Models with Multimodal Contextual Reasoning

Recent progress in foundation models has shifted toward agentic behavior involving multi-step reasoning and tool use. However, open-source efforts largely focus on text-dominant settings, leaving long-horizon multimodal tasks underexplored. This gap is evident in video tasks requiring sustained temporal understanding and iterative interaction. We present InternVideo3, a framework enhancing these capabilities via Multimodal Contextual Reasoning (MCR). MCR treats understanding as a closed-loop process over a shared, evolving context containing observations, instructions, reasoning, tool actions, and memory. This frames long-video understanding as evidence accumulation and verification. To ensure efficiency, we introduce Multimodal Multi-head Latent Attention (M^2LA), a token-preserving reparameterization compressing KV-cache states while retaining the full token stream. Our staged training includes continued pretraining, short-to-long supervised fine-tuning, rule-based reinforcement learning, and on-policy distillation. Experiments show InternVideo3 achieves strong performance on benchmarks like Video-MME, MLVU, and EgoSchema. We further instantiate the model as a video agent with retrieval tools, demonstrating robust evidence-grounded behavior. Our results suggest that efficient context handling and closed-loop reasoning are vital for adapting open multimodal models toward long-horizon visually grounded agency.

cs.CV↗

arXiv2026

ViCuR: Visual Cues as Recoverable Privilege for Multimodal On-Policy Distillation

On-policy distillation (OPD) improves reasoning by training a student on trajectories sampled from its own policy under supervision from a teacher. In multimodal reasoning, a common extension is to use a privileged teacher that observes training-time-only signals such as reference answers or rationales. However, such answer-side privilege creates a train-test mismatch: the teacher's supervision may depend on signals unavailable to the student, encouraging shortcut imitation rather than visually grounded reasoning. We propose ViCuR, a visually grounded privileged-teacher distillation framework that replaces answer-side privilege with visual cues (query-related evidence in the input). Because these cues are derived from the same visual input available at inference, their evidence is recoverable by the student. To support this, ViCuR introduces a lightweight cue recovery module that uses dedicated sink-token cross-attention during prefill to aggregate task-relevant visual evidence into an internal representation, without changing the inference interface or requiring auxiliary cue-generation losses. Across seven benchmarks with Qwen3-VL-2B and 8B students, ViCuR consistently improves over answer-based on-policy self-distillation by +1.19 and +1.24 on overall average performance. It also extends naturally to stronger-teacher OPD, surpassing OPD baselines by +0.64 and +1.08, with consistent out-of-domain gains at the 8B scale. These results show that, in multimodal on-policy distillation, the design of teacher privilege is as important as teacher strength.

cs.CV↗

arXiv2026

Imagine Before You Predict: Interleaved Latent Visual Reasoning for Video Event Prediction

Video event prediction (VEP) requires models to infer unobserved future states from partial video evidence. Existing video MLLMs usually verbalize intermediate future reasoning in text space: once visual evidence is verbalized, fine-grained motion, geometry, and interaction cues can be lost, leading to plausible but visually ungrounded hallucinations. We introduce Future-L1, an interleaved latent visual reasoning framework that lets an MLLM alternate between language tokens and continuous latent visual spans during autoregressive decoding. To train this capability, we construct Future-L1-50K by selecting examples where future visual hints help prediction and align latent states to future-frame embeddings, then further optimize sampled latent trajectories with LA-DAPO, a latent-aware RL objective with outcome-contrastive and temporal-diversity rewards. Future-L1 achieves new state-of-the-art results on both benchmarks: on FutureBench, it improves Qwen3-VL-8B from 61.0 to 85.4 and exceeds the previous best Video-CoE by 10.4 points; on TwiFF-Bench, it improves the average score from 2.44 to 3.04. These results suggest that future-oriented video reasoning benefits from preserving intermediate visual semantics in latent space rather than translating every reasoning step into text.

cs.CV↗

arXiv2026

Intern-S1-Pro: Scientific Multimodal Foundation Model at Trillion Scale

We introduce Intern-S1-Pro, the first one-trillion-parameter scientific multimodal foundation model. Scaling to this unprecedented size, the model delivers a comprehensive enhancement across both general and scientific domains. Beyond stronger reasoning and image-text understanding capabilities, its intelligence is augmented with advanced agent capabilities. Simultaneously, its scientific expertise has been vastly expanded to master over 100 specialized tasks across critical science fields, including chemistry, materials, life sciences, and earth sciences. Achieving this massive scale is made possible by the robust infrastructure support of XTuner and LMDeploy, which facilitates highly efficient Reinforcement Learning (RL) training at the 1-trillion parameter level while ensuring strict precision consistency between training and inference. By seamlessly integrating these advancements, Intern-S1-Pro further fortifies the fusion of general and specialized intelligence, working as a Specializable Generalist, demonstrating its position in the top tier of open-source models for general capabilities, while outperforming proprietary models in the depth of specialized scientific tasks.

cs.LG↗

arXiv2026

Video-o3: Native Interleaved Clue Seeking for Long Video Multi-Hop Reasoning

Existing multimodal large language models for long-video understanding predominantly rely on uniform sampling and single-turn inference, limiting their ability to identify sparse yet critical evidence amid extensive redundancy. We introduce Video-o3, a novel framework that supports iterative discovery of salient visual clues, fine-grained inspection of key segments, and adaptive termination once sufficient evidence is acquired. Technically, we address two core challenges in interleaved tool invocation. First, to mitigate attention dispersion induced by the heterogeneity of reasoning and tool-calling, we propose Task-Decoupled Attention Masking, which isolates per-step concentration while preserving shared global context. Second, to control context length growth in multi-turn interactions, we introduce a Verifiable Trajectory-Guided Reward that balances exploration coverage with reasoning efficiency. To support training at scale, we further develop a data synthesis pipeline and construct Seeker-173K, comprising 173K high-quality tool-interaction trajectories for effective supervised and reinforcement learning. Extensive experiments show that Video-o3 substantially outperforms state-of-the-art methods, achieving 72.1% accuracy on MLVU and 46.5% on Video-Holmes. These results demonstrate Video-o3's strong multi-hop evidence-seeking and reasoning capabilities, and validate the effectiveness of native tool invocation in long-video scenarios.

cs.CV↗

arXiv2026

KiDS-Legacy: WIMP dark matter constraints from the cross-correlation of weak lensing and Fermi-LAT gamma rays

Dark matter dominates the matter content of the Universe, and its properties can be constrained through large-scale structure probes such as the cross-correlation between the unresolved gamma-ray background (UGRB) and weak gravitational lensing. We analysed 15 years of Fermi-LAT data, constructing UGRB intensity maps in ten energy bins (0.5-1000 GeV), and cross-correlated them with KiDS-Legacy shear in six tomographic bins. The measurements were performed using angular power spectra estimated with the pseudo-$C_\ell$ method. No significant cross-correlation is found. Based on this non-detection, we present 95% upper bounds on the weakly interacting massive particle (WIMP) decay rate $\Gamma_{\rm dec}$ and velocity-averaged annihilation cross-section $\langle\sigma_{\rm ann} v\rangle$ as functions of mass. We compare our results with bounds from other cosmological tracers and from local probes, and found them to be complementary, particularly at low masses ($\rm GeV/TeV$). In addition, using a Euclid-like lensing survey cross-correlated with Fermi-LAT, we forecast $\sim$2 times tighter limits, highlighting the potential of forthcoming data to strengthen constraints on dark matter annihilation and decay.

astro-ph.CO↗

arXiv2026

Diagnosing the Effects of Spectroscopic Training Set Imperfection on Photometric Redshift Performance

Most LSST extragalactic science will rely on photometric redshifts (photo-$z$) to extract distance information for the galaxies. However, an incomplete or non-representative training set can introduce bias into photo-$z$ estimation. It is necessary to understand how various forms of training set imperfection, such as incompleteness and non-trivial spectroscopic target selection, affect photo-$z$ estimation algorithms, and to identify metrics best-suited to quantify the impact. This work aims to systematically study metrics for diagnosing how various photo-$z$ methods react to certain types of training set incompleteness and non-representativeness. We use methods available through the open-source Python library Redshift Assessment Infrastructure Layers (RAIL) to systematically test the algorithms CMNN, GPz, FlexZBoost, and PZFlow on mock training data degraded in accordance with several existing spectroscopic sky surveys, as well as under conditions of inverse redshift incompleteness, which approximately mimics observed patterns of incompleteness at high redshift. We employ the algorithm TrainZ as a control. Finally, we quantify photo-$z$ algorithm performance using a variety of statistical metrics implemented externally to RAIL. We determine that the Kullback-Liebler Divergence, Wasserstein Distance, and Probability Integral Transform are particularly informative metrics with which to assess the impact of training set imperfection on algorithmic performance. We also find that inverse redshift incompleteness effects alone lack the complexity to realistically represent anticipated training data.

astro-ph.IM↗

arXiv2025

KiDS-Legacy: Constraints on Horndeski gravity from weak lensing combined with galaxy clustering and cosmic microwave background anisotropies

We present constraints on modified gravity from a cosmic shear analysis of the final data release of the Kilo-Degree Survey (KiDS-Legacy) in combination with DESI measurements of baryon acoustic oscillations, eBOSS observations of redshift space distortions, and cosmic microwave background anisotropies from Planck. We study the Horndeski class of modified gravity models in an effective field theory framework employing a parameterisation that satisfies stability conditions by construction and, for the first time, present a cosmological analysis in this inherently stable parameter basis. Cosmic shear constrains the Horndeski parameter space significantly, matching or surpassing the CMB contribution. Adopting the de-mixed kinetic term of the scalar field perturbation, $D_{\rm kin}$, and the deviation of the Planck mass from its fiducial value, $\Delta M_*^2\equiv M_*^2-1$, as model parameters, we constrain their present values to be $\Delta \hat{M}_*^2=0.32^{+0.07}_{-0.21}$ and $\hat{D}_{\rm kin} = 3.74^{+0.69}_{-1.92}$, which deviate from general relativity at $1.5\sigma$ and $1.9\sigma$, respectively. We derive constraints on the structure growth parameter $S_8=0.813^{+0.008}_{-0.011}$, which is compatible with the $\Lambda$CDM constraint at $0.54\sigma$. We obtain the deviation of the effective Newtonian coupling from the GR value as $\Delta \mu_{\infty,{\rm eff}}=0.066\pm0.023$, corresponding to a $2.9\sigma$ significance. Although modified gravity provides a slightly better fit to the data, a model comparison shows only a weak preference for modified gravity at the $1.4\sigma$ level. When adopting a dynamical dark energy model of the background cosmology, the inferred modified gravity parameter constraints are stable with respect to a $\Lambda$CDM background, while a mild preference at $1.57\sigma$ for dynamical dark energy remains.

astro-ph.CO↗

arXiv2025

KiDS-Legacy: Constraining dark energy, neutrino mass, and curvature

We constrained minimally extended cosmological models with the cosmic shear analysis of the final data release from the Kilo-Degree Survey (KiDS-Legacy) in combination with external probes. Due to the consistency of the KiDS-Legacy analysis with the cosmic microwave background (CMB), we could combine these datasets reliably for the first time. Additionally, we used CMB lensing, galaxy redshift-space distortions, and baryon acoustic oscillations. We assessed, in turn, the effects of spatial curvature, varying neutrino masses, and an evolving dark energy component on cosmological constraints from KiDS-Legacy alone and from KiDS-Legacy combined with external probes. We find KiDS-Legacy to be consistent with the fiducial flat $\Lambda$-cold dark matter ($\Lambda$CDM) analysis with $c^2 \sum m_\nu\leq 1.5\,$eV, $w_0 = -1.0\pm 0.7$, and $w_a = -1.3^{+1.9}_{-2.0}$ while $\Omega_K = 0.08^{+0.16}_{-0.17}$ (1$\sigma$ bounds) with an almost equal goodness of fit. The $w_0w_a$CDM model is not a significant improvement over $\Lambda$CDM when cosmic shear and CMB lensing are combined, yielding a Bayes factor $B = 0.07$. If all probes are combined, however, $B$ increases to 2.73, corresponding to a $2.6\sigma$ suspiciousness tension. The constraint on $S_8 = \sigma_8\sqrt{\Omega_\mathrm{m}/0.3}$ is robust to opening up the parameter space for cosmic shear. Adding all external datasets to KiDS-Legacy, we find $S_8 = 0.816 \pm 0.006$ in $\Lambda$CDM and $S_8 = 0.837 \pm 0.008$ in $w_0 w_a$CDM for all probes combined.

astro-ph.CO↗

arXiv2025

InternVideo-Next: Towards General Video Foundation Models without Video-Text Supervision

Large-scale video-text pretraining achieves strong performance but depends on noisy, synthetic captions with limited semantic coverage, often overlooking implicit world knowledge such as object motion, 3D geometry, and physical cues. In contrast, masked video modeling (MVM) directly exploits spatiotemporal structures but trails text-supervised methods on general tasks. We find this gap arises from overlooked architectural issues: pixel-level reconstruction struggles with convergence and its low-level requirement often conflicts with semantics, while latent prediction often encourages shortcut learning. To address these, we disentangle the traditional encoder-decoder design into an Encoder-Predictor-Decoder (EPD) framework, where the predictor acts as a latent world model, and propose InternVideo-Next, a two-stage pretraining scheme that builds a semantically consistent yet detail-preserving latent space for this world model. First, conventional linear decoder in pixel MVM enforces the predictor output latent to be linearly projected to, thus separable in pixel space, causing the conflict with semantic abstraction. Our Stage 1 proposes a conditional diffusion decoder and injects reliable image-level semantic priors to enhance semantics and convergence, thus bridging pixel-level fidelity with high-level semantic abstraction. Stage 2 further learns world knowledge by predicting frozen Stage 1 targets within this space, mitigating shortcut learning. Trained on public, unlabeled videos, InternVideo-Next achieves state-of-the-art results across benchmarks and provides a scalable path toward general video representation learning.

cs.CV↗

arXiv2025

EchoVLA: Robotic Vision-Language-Action Model with Synergistic Declarative Memory for Mobile Manipulation

Recent progress in Vision-Language-Action (VLA) models has enabled embodied agents to interpret multimodal instructions and perform complex tasks. However, existing VLAs are mostly confined to short-horizon, table-top manipulation, lacking the memory and reasoning capability required for mobile manipulation, where agents must coordinate navigation and manipulation under changing spatial contexts. In this work, we present EchoVLA, a memory-aware VLA model for mobile manipulation. EchoVLA incorporates a synergistic declarative memory inspired by the human brain, consisting of a scene memory that maintains a collection of spatial-semantic maps and an episodic memory that stores task-level experiences with multimodal contextual features. The two memories are individually stored, updated, and retrieved based on current observations, task history, and instructions, and their retrieved representations are fused via coarse- and fine-grained attention to guide base-arm diffusion policies. To support large-scale training, we further introduce MoMani, an automated benchmark that generates expert-level trajectories through multimodal large language model (MLLM)-guided planning and feedback-driven refinement, supplemented with real-robot demonstrations. Comprehensive simulated and real-world results demonstrate that EchoVLA substantially improves overall performance, e.g., it achieves the highest success rates of 0.52 on manipulation/navigation tasks and 0.31 on mobile manipulation tasks in simulation, exceeding the strong baseline $\pi_{0.5}$ by +0.20 and +0.11, respectively.

cs.RO↗

arXiv2025

VideoTG-R1: Boosting Video Temporal Grounding via Curriculum Reinforcement Learning on Reflected Boundary Annotations

Video temporal grounding (VTG) aims to locate precise segments in videos based on language queries, which is a fundamental challenge in video understanding. While recent Multimodal Large Language Models (MLLMs) have shown promise in tackling VTG through reinforcement learning (RL), they overlook the challenges arising from both the quality and difficulty of training samples. (1) Partially annotated samples. Many samples contain relevant segments beyond the annotated interval, introducing ambiguous supervision. (2) Hard-to-ground samples. Samples with poor zero-shot performance produce consistently low and indistinguishable rewards during RL training, exhibiting no clear preference among multiple outputs and thus hindering learning efficiency. To address these challenges, we propose VideoTG-R1, a novel curriculum RL framework with reflected boundary annotations, enabling data-efficient training. Specifically, we propose a Boundary Reflection Agent that utilizes MLLMs to predict query-relevant timestamps outside the annotated intervals, allowing us to identify and filter out partially annotated samples, thereby reducing ambiguity. Furthermore, we introduce a Difficulty Estimation Agent to assess the training difficulty of each sample and design a curriculum RL strategy that dynamically masks the videos of hard-to-ground samples according to the training steps, easing the training difficulty and providing clearer preference. Experiments on the VTG and grounded VideoQA tasks demonstrate the effectiveness of our method. Remarkably, with only 10% of the training samples and 21% of the computational budget, VideoTG-R1 outperforms full-data counterparts under both group relative policy optimization (GRPO) and supervised fine-tuning (SFT). The code is available at https://github.com/ldong1111/VideoTG-R1.

cs.CV↗

arXiv2025

DSI-Bench: A Benchmark for Dynamic Spatial Intelligence

Reasoning about dynamic spatial relationships is essential, as both observers and objects often move simultaneously. Although vision-language models (VLMs) and visual expertise models excel in 2D tasks and static scenarios, their ability to fully understand dynamic 3D scenarios remains limited. We introduce Dynamic Spatial Intelligence and propose DSI-Bench, a benchmark with nearly 1,000 dynamic videos and over 1,700 manually annotated questions covering nine decoupled motion patterns of observers and objects. Spatially and temporally symmetric designs reduce biases and enable systematic evaluation of models' reasoning about self-motion and object motion. Our evaluation of 14 VLMs and expert models reveals key limitations: models often conflate observer and object motion, exhibit semantic biases, and fail to accurately infer relative relationships in dynamic scenarios. Our DSI-Bench provides valuable findings and insights about the future development of general and expertise models with dynamic spatial intelligence.

cs.CV↗

arXiv2025

ExpVid: A Benchmark for Experiment Video Understanding & Reasoning

Multimodal Large Language Models (MLLMs) hold promise for accelerating scientific discovery by interpreting complex experimental procedures. However, their true capabilities are poorly understood, as existing benchmarks neglect the fine-grained and long-horizon nature of authentic laboratory work, especially in wet-lab settings. To bridge this gap, we introduce ExpVid, the first benchmark designed to systematically evaluate MLLMs on scientific experiment videos. Curated from peer-reviewed video publications, ExpVid features a new three-level task hierarchy that mirrors the scientific process: (1) Fine-grained Perception of tools, materials, and actions; (2) Procedural Understanding of step order and completeness; and (3) Scientific Reasoning that connects the full experiment to its published conclusions. Our vision-centric annotation pipeline, combining automated generation with multi-disciplinary expert validation, ensures that tasks require visual grounding. We evaluate 19 leading MLLMs on ExpVid and find that while they excel at coarse-grained recognition, they struggle with disambiguating fine details, tracking state changes over time, and linking experimental procedures to scientific outcomes. Our results reveal a notable performance gap between proprietary and open-source models, particularly in high-order reasoning. ExpVid not only provides a diagnostic tool but also charts a roadmap for developing MLLMs capable of becoming trustworthy partners in scientific experimentation.

cs.CV↗