arXiv subjects

Haochen Wang

Publications and source records attributed to Haochen Wang.

At least 37 records · Page 2Linked to original sources

arXiv2026

SAMTok: Representing Any Mask with Two Words

Pixel-wise capabilities are essential for building interactive intelligent systems. However, pixel-wise multi-modal LLMs (MLLMs) remain difficult to scale due to complex region-level encoders, specialized segmentation decoders, and incompatible training objectives. To address these challenges, we present SAMTok, a discrete mask tokenizer that converts any region mask into two special tokens and reconstructs the mask using these tokens with high fidelity. By treating masks as new language tokens, SAMTok enables base MLLMs (such as the QwenVL series) to learn pixel-wise capabilities through standard next-token prediction and simple reinforcement learning, without architectural modifications and specialized loss design. SAMTok builds on SAM2 and is trained on 209M diverse masks using a mask encoder and residual vector quantizer to produce discrete, compact, and information-rich tokens. With 5M SAMTok-formatted mask understanding and generation data samples, QwenVL-SAMTok attains state-of-the-art or comparable results on region captioning, region VQA, grounded conversation, referring segmentation, scene graph parsing, and multi-round interactive segmentation. We further introduce a textual answer-matching reward that enables efficient reinforcement learning for mask generation, delivering substantial improvements on GRES and GCG benchmarks. Our results demonstrate a scalable and straightforward paradigm for equipping MLLMs with strong pixel-wise capabilities. Our code and models are available.

cs.CV↗

arXiv2026

DRIVE: Distributional and Retrieval-Augmented Bidding with Value Evaluation

Auto-bidding is a core component of real-time advertising systems, where decisions must optimize long-term performance under budget and cost constraints, while online exploration is prohibitively risky. Offline reinforcement learning and, more recently, Transformer-based sequence modeling have shown promise for learning bidding policies from logged data, but their unimodal and purely parametric formulations often collapse multiple effective bidding strategies into suboptimal averaged actions and perform unreliably under sparse or long-tail traffic. To mitigate these limitations, we propose DRIVE (Distributional and Retrieval-Augmented Bidding with Value Evaluation), a unified Transformer-based framework that decouples candidate action generation from decision making for offline auto-bidding. DRIVE combines distributional action modeling, retrieval-augmented candidate generation from high-quality historical decisions, and value-based evaluation to select the most promising bid at inference time. Extensive experiments on AuctionNet and additional offline reinforcement learning benchmarks demonstrate that DRIVE consistently improves bidding performance and generalizes well across multiple Transformer-based methods.

cs.LG↗

arXiv2026

Watch, Remember, Reason: Human-View Video Understanding with MLLMs

Video understanding is being rapidly transformed by multimodal large language models (MLLMs), as research moves from short clips to long, multimodal, and knowledge-intensive video scenarios. These scenarios require models to handle sparse evidence, long-range dependencies, multimodal alignment, and reliable inference under limited computational budgets. This work presents a human-view perspective on LLM-based video understanding, organized around three functional abilities: watching, remembering, and reasoning. Rather than treating video tasks as isolated benchmarks, this view provides a unified structure for analyzing how video MLLMs acquire evidence, preserve context, and produce grounded outputs. We introduce a formulation that characterizes video understanding systems by their perceptual representations, memory states, reasoning traces, and final predictions. Based on this formulation, we identify challenges in spatio-temporal perception, efficient long-video processing, memory modeling, streaming understanding, and faithful reasoning. Representative methods are organized by their roles in video MLLM systems. Watching covers fine-grained, comprehensive, audio-visual, and efficient perception. Remembering includes offline and streaming memory, while reasoning covers text-only reasoning and thinking with videos. We further examine application domains such as egocentric, sports, instructional, medical, and narrative videos, and cover training datasets and evaluation benchmarks across task types, supervision formats, modalities, and capability dimensions. Finally, we outline open problems and future directions for scalable, memory-aware, and evidence-grounded video intelligence. Related works will be continuously traced at https://github.com/marinero4972/Awesome-HumanView-VideoUnderstanding.

cs.CV↗

arXiv2026

ClaimDiff-RL: Fine-Grained Caption Reinforcement Learning through Visual Claim Comparison

Long-form image captioning exposes a reward granularity problem in RL: captions are judged as whole sequences, while the important errors occur at the level of individual visual claims. A good dense caption should be both faithful and informative, avoiding hallucination without omitting salient details. Yet pairwise preferences, reference-based metrics, and holistic scalar rewards compress these local errors into a single sequence-level signal, obscuring the tradeoff between factuality and coverage. We introduce ClaimDiff-RL, a framework that uses reference-conditioned atomic claim differences as the reward unit for caption RL. Given an image, an actor caption, and a reference caption, a multimodal judge enumerates visually grounded differences, verifies each difference against the image, assigns open-vocabulary error types and severity levels, and produces per-difference statistics for reward composition. This makes hallucinated claims and omitted salient facts separately measurable and tunable. Experiments show that holistic scalar rewards can reduce hallucination by increasing missing facts, while ClaimDiff-RL exposes this faithfulness and coverage tradeoff and enables more balanced operating points. On a 160-image human-labeled diagnostic benchmark, public captioning benchmarks, and VQA benchmarks, ClaimDiff-RL improves the hallucination--missing-fact balance, preserves general capability, and even surpasses Gemini-3-Pro-Preview on several fine-grained Capability dimensions such as object counting, spatial relations, and scene recognition. These results suggest that typed, verifiable claim differences are an effective reward unit for fine-grained and diagnosable caption RL.

cs.LG↗

arXiv2026

Toward a unified data-driven turbulence model through multi-objective learning

Turbulence remains one of the last unresolved problems of classical physics and a major bottleneck to accurate flow prediction in climate, aerospace, and energy systems. Industrial simulations therefore rely on averaged representations of turbulence, which often struggle to predict flows governed by multiple interacting mechanisms. We present a unified, data-driven turbulence modeling framework designed to learn robustly from sparse, indirect observations across diverse flow regimes. The framework embeds physical consistency into a flexible, frame-invariant closure, automatically selects representative training cases based on similarity of flow-feature distributions, and learns a single, unified model through a multi-objective ensemble strategy that balances competing objectives across flows and quantities of interest. The resulting unified foundation model adapts seamlessly across regimes without manual intervention. It outperforms existing turbulence models across a broad spectrum of canonical flows and maintains improved performance in complex three-dimensional configurations of industrial relevance, including a gas turbine diffuser, a generic car, and a generic aircraft. When application-specific accuracy is required, the framework further enables specialist models through additive fine-tuning on targeted flow datasets. The results demonstrate the feasibility of a deployable and generalized turbulence modeling approach that unifies multiple flow mechanisms within a single architecture for a broad range of natural and industrial flows.

physics.flu-dyn↗

arXiv2026

VGR: Visual Grounded Reasoning

In the field of multimodal chain-of-thought (CoT) reasoning, existing approaches predominantly rely on reasoning on pure language space, which inherently suffers from language bias and is largely confined to math or science domains. This narrow focus limits their ability to handle complex visual reasoning tasks that demand comprehensive understanding of image details. To address these limitations, this paper introduces VGR, a novel reasoning multimodal large language model (MLLM) with enhanced fine-grained visual perception capabilities. Unlike traditional MLLMs that answer the question or reasoning solely on the language space, our VGR first detects relevant regions that may help to solve problems, and then provides precise answers based on replayed image regions. To achieve this, we conduct a large-scale SFT dataset called VGR -SFT that contains reasoning data with mixed vision grounding and language deduction. The inference pipeline of VGR allows the model to choose bounding boxes for visual reference and a replay stage is introduced to integrates the corresponding regions into the reasoning process, enhancing multimodel comprehension. Experiments on the LLaVA-NeXT-7B baseline show that VGR achieves superior performance on multi-modal benchmarks requiring comprehensive image detail understanding. Compared to the baseline, VGR uses only 30\% of the image token count while delivering scores of +4.1 on MMStar, +7.1 on AI2D, and a +12.9 improvement on ChartQA.

cs.CV↗

arXiv2026

Fast radio burst dispersion is an unbiased tracer of matter on large scales

The dispersion of fast radio bursts (FRBs) measures the column density of free electrons, tracing the diffuse ionized gas that contains more than $90\%$ of all baryons. On linear scales the FRB dispersion field is an approximately unbiased tracer of the matter distribution, an idea long assumed in the FRB large-scale structure literature and recently formalized by Zhou and Zhang [arXiv:2510.11022]. This follows from baryon-mass conservation, which forces the total baryon field to have unit linear bias, with dispersion inheriting this bias up to small corrections from the stellar and neutral-gas components. We show these corrections can be bounded at the percent level using existing galaxy and 21 cm surveys, and confirm with the FLAMINGO hydrodynamical simulations that the electron bias varies at the percent level across a wide range of feedback prescriptions. The dispersion-galaxy cross-power spectrum at linear scales directly constrains $B_8 \equiv σ_8(Ω_b/0.05)^{1/2}$, a baryonic analog of $S_8$, independently of feedback physics. Because most of the per-object variance in dispersion is cosmological signal rather than noise, $\sim\!10^5$ localized FRBs can match the statistical power of $\sim\!10^8$ weak-lensing galaxy shape measurements. FRB dispersion thus joins weak lensing and redshift-space distortions as a new unbiased tracer of matter on large scales.

astro-ph.CO↗

arXiv2026

Constraining Gas Mass Fractions in Galaxy Groups and Clusters with the First CHIME/FRB Outrigger

In recent years, localized fast radio bursts (FRBs) have emerged as a powerful tool to study the structure of the baryonic matter in the universe. Their dispersion measures (DMs) scale linearly with electron density independent of gas temperature, making them particularly well suited to studying the intragroup medium (IGrM), where traditional probes such as X-ray emission and the SZ effect are weak. Evidence suggests that the gas in group mass halos ($M_{500}$ ~ $10^{13}$ -- $10^{14}$ M$_\odot$) is strongly affected by galactic feedback, causing deviations from cluster scaling relations. Three FRBs from the first CHIME/FRB Outrigger sample come from host galaxies found within or behind galaxy clusters and groups. We estimate the DM contribution of each ICM/IGrM by integrating different halo density profiles, accounting for uncertainties in halo mass and the host galaxy line of sight distance. For the more massive halos, predicted cluster DMs agree with the extragalactic DM budget. One burst, FRB 20230703A, intersects three groups yet has a low extragalactic DM. By comparing model predictions with the measured DM, we constrain the gas mass fraction $f_g(R)$ in these halos. Comparing with published $M$--$f_g$ relations, we find consistency with recent eROSITA results at $R_{500}$ and mild tension at $R_{200}$ and with earlier X-ray--based relations. As CHIME/FRB Outriggers build a large catalog of localized FRBs, many additional sightlines through groups and clusters will be obtained. These will enable systematic tests of intragroup and intracluster gas properties and sharpen constraints on the distribution of baryons in massive halos.

astro-ph.GA↗

arXiv2026

Unified Multimodal Models as Auto-Encoders

Image-to-text (I2T) understanding and text-to-image (T2I) generation are two fundamental, important yet traditionally isolated multimodal tasks. Despite their intrinsic connection, existing approaches typically optimize them independently, missing the opportunity for mutual enhancement. In this paper, we argue that the both tasks can be connected under a shared Auto-Encoder perspective, where text serves as the intermediate latent representation bridging the two directions - encoding images into textual semantics (I2T) and decoding text back into images (T2I). Our key insight is that if the encoder truly "understands" the image, it should capture all essential structure, and if the decoder truly "understands" the text, it should recover that structure faithfully. Building upon this principle, we propose Unified-GRPO, a post-training method based on reinforcement learning that jointly optimizes both modules through reconstructive rewards, maximizing the semantic consistency between the input and the generated images. Under this reconstruction objective, the encoder is encouraged to extract as much accurate and comprehensive semantic information from the input image to maximize reconstruction quality, while the decoder is simultaneously optimized to generate conditioned on the encoder's prior, enabling a self-evolving improvement. Empirically, we find that using text as the intermediate representation and training under a reconstructive RL paradigm effectively benefits both I2T and T2I. The I2T module gains stronger fine-grained visual perception, such as small-object recognition, grounding, etc, while its dense embeddings and language priors, in turn, provide richer semantic signals that improve T2I fidelity and complex instruction following. These results demonstrate that the reconstructive RL establishes a mutually reinforcing cross-modal synergy within the auto-encoding framework.

cs.CV↗

arXiv2026

Detection of the Cosmological 21 cm Signal in Auto-correlation at z ~ 1 with the Canadian Hydrogen Intensity Mapping Experiment

We present the first detection of the cosmological 21 cm intensity mapping signal in auto-correlation at z ~ 1 with the Canadian Hydrogen Intensity Mapping Experiment (CHIME). Using 94 nights of observation, we have measured the 21 cm auto-power spectrum over a frequency range from 608.2 MHz to 707.8 MHz (z = 1.34 to 1.01) at 0.4 h Mpc^-1 < k < 1.5 h Mpc^-1, with a detection significance of 12.5 sigma. Our analysis employs significant improvements to the CHIME data processing pipeline compared to previous work, including novel radio frequency interference (RFI) detection and masking algorithms, achromatic beamforming techniques, and foreground filtering before time averaging to minimize spectral leakage. We establish the robustness and reliability of our detection through a comprehensive suite of validation tests. We also measure the 21 cm signal in two independent sub-bands centered at z ~ 1.08 and z ~ 1.24 with detection significance of 8.7 sigma and 9.2 sigma, respectively. We briefly discuss the theoretical interpretation of these measurements in terms of a power spectrum model, deferring the details to a companion paper. This auto-power spectrum detection demonstrates CHIME's capability to probe large-scale structure through 21 cm intensity mapping without reliance on external galaxy surveys.

astro-ph.CO↗

arXiv2026

Interpretation of 21 cm Auto Power Spectrum Measurement at $z\sim 1$ by the Canadian Hydrogen Intensity Mapping Experiment

Observations with the Canadian Hydrogen Intensity Mapping Experiment (CHIME) have been used to measure the 21 cm intensity mapping auto power spectrum, at $z\sim 1$, over a frequency range from 608.2 MHz to 707.8 MHz at wavenumbers $0.4~h~{\rm Mpc}^{-1} \lesssim k \lesssim 1.5~h~{\rm Mpc}^{-1}$. In this paper, we present the results of two different approaches to interpreting this measurement. In the first approach, we use a parametric power spectrum model to constrain an amplitude parameter, defined as $\mathcal{A}^2_{\rm HI} \equiv 10^6 Ω_{\rm HI}^2(b^2_{\rm HI}+\langle f μ^2\rangle)^2$, where $Ω_{\rm HI}$ is the cosmological density parameter for atomic hydrogen ($\rm HI$), $b_{\rm HI}$ is the linear bias for $\rm HI$, and $\langle f μ^2\rangle$ incorporates the dominant large-scale impact of redshift-space distortions on the angle-averaged power spectrum. Imposing an additional prior on either $Ω_{\rm HI}$ or $b_{\rm HI}$, based on values in the literature, allows us to break the pairwise degeneracy between those two parameters. In the second approach, we compare CHIME's measurement with predictions for the power spectrum of $\rm HI$ from the IllustrisTNG simulations, finding that the measurement disagrees with the TNG100 run at $3.1σ$ and the TNG300 run at $4.0σ$. This disagreement is most likely attributable to the strength of nonlinear redshift-space clustering of $\rm HI$ in the simulations, rather than the total abundance of $\rm HI$, and invites further investigation of the physical processes in the simulations that determine the behavior of $\rm HI$ at nonlinear scales. These results exemplify the ability of 21 cm intensity mapping to provide astrophysical information using measurements at nonlinear scales.

astro-ph.CO↗

arXiv2026

Open-o3-Video: Grounded Video Reasoning with Explicit Spatio-Temporal Evidence

Most video reasoning models only generate textual reasoning traces without indicating when and where key evidence appears. Recent models such as OpenAI-o3 have sparked wide interest in evidence-centered reasoning for images, yet extending this ability to videos is more challenging due to the need for joint temporal tracking and spatial localization across dynamic scenes. We introduce Open-o3-Video, a non-agent framework that integrates explicit spatio-temporal evidence into video reasoning by highlighting key timestamps, objects, and bounding boxes, making the reasoning process traceable and verifiable. To enable this capability, we first construct high-quality datasets STGR that provide unified spatio-temporal supervision, which is absent in existing resources. We further adopt a cold-start reinforcement learning strategy with specially designed rewards that jointly encourage answer accuracy, temporal alignment, and spatial precision. On the V-STAR benchmark, Open-o3-Video achieves state-of-the-art performance, improving mAM by 14.4% and mLGM by 24.2% over the Qwen2.5-VL baseline, and shows consistent gains across a range of video understanding benchmarks. Beyond accuracy, the grounded reasoning traces produced by Open-o3-Video support confidence-aware test-time scaling, improving answer reliability.

cs.CV↗

arXiv2026

Traceable Evidence Enhanced Visual Grounded Reasoning: Evaluation and Methodology

Models like OpenAI-o3 pioneer visual grounded reasoning by dynamically referencing visual regions, just like human "thinking with images". However, no benchmark exists to evaluate these capabilities holistically. To bridge this gap, we propose TreeBench (Traceable Evidence Evaluation Benchmark), a diagnostic benchmark built on three principles: (1) focused visual perception of subtle targets in complex scenes, (2) traceable evidence via bounding box evaluation, and (3) second-order reasoning to test object interactions and spatial hierarchies beyond simple object localization. Prioritizing images with dense objects, we initially sample 1K high-quality images from SA-1B, and incorporate eight LMM experts to manually annotate questions, candidate options, and answers for each image. After three stages of quality control, TreeBench consists of 405 challenging visual question-answering pairs, even the most advanced models struggle with this benchmark, where none of them reach 60% accuracy, e.g., OpenAI-o3 scores only 54.87. Furthermore, we introduce TreeVGR (Traceable Evidence Enhanced Visual Grounded Reasoning), a training paradigm to supervise localization and reasoning jointly with reinforcement learning, enabling accurate localizations and explainable reasoning pathways. Initialized from Qwen2.5-VL-7B, it improves V* Bench (+16.8), MME-RealWorld (+12.6), and TreeBench (+13.4), proving traceability is key to advancing vision-grounded reasoning. The code is available at https://github.com/Haochen-Wang409/TreeVGR.

cs.CV↗

arXiv2026

Grasp Any Region: Towards Precise, Contextual Pixel Understanding for Multimodal LLMs

While Multimodal Large Language Models (MLLMs) excel at holistic understanding, they struggle in capturing the dense world with complex scenes, requiring fine-grained analysis of intricate details and object inter-relationships. Region-level MLLMs have been a promising step. However, previous attempts are generally optimized to understand given regions in isolation, neglecting crucial global contexts. To address this, we introduce Grasp Any Region (GAR) for comprehen- sive region-level visual understanding. Empowered by an effective RoI-aligned feature replay technique, GAR supports (1) precise perception by leveraging necessary global contexts, and (2) modeling interactions between multiple prompts. Together, it then naturally achieves (3) advanced compositional reasoning to answer specific free-form questions about any region, shifting the paradigm from passive description to active dialogue. Moreover, we construct GAR-Bench, which not only provides a more accurate evaluation of single-region comprehension, but also, more importantly, measures interactions and complex reasoning across multiple regions. Extensive experiments have demonstrated that GAR-1B not only maintains the state-of-the-art captioning capabilities, e.g., outperforming DAM-3B +4.5 on DLC-Bench, but also excels at modeling relationships between multiple prompts with advanced comprehension capabilities, even surpassing InternVL3-78B on GAR-Bench-VQA. More importantly, our zero-shot GAR-8B even outperforms in-domain VideoRefer-7B on VideoRefer-BenchQ, indicating its strong capabilities can be easily transferred to videos.

cs.CV↗

arXiv2026

Strain effects on $n$-type doping in AlN

Controllable doping in AlN and its alloys is essential for deep-ultraviolet light sources. Ionization energies for donors in AlN ($\mathrm{Si_{Al}}$, $\mathrm{S_N}$, $\mathrm{Se_N}$) are high. We report first-principles calculations demonstrating that strain engineering can result in a reduction in ionization energies. The donor levels for $\mathrm{S_N}$ and $\mathrm{Se_N}$ shift closer to the conduction-band minimum (CBM) under in-plane tensile strains, driven by a downward shift of the CBM. The most widely used donor, $\mathrm{Si_{Al}}$, forms a $DX$ center in AlN. We find that a 2.5% in-plane tensile strain (which would be induced by pseudomorphic growth on GaN in experiment) shifts the ($+/-$) transition level from 271 meV to 98 meV below the CBM, which would enhance the electron concentration by three orders of magnitude. These results demonstrate that strain engineering offers an effective route to enhance doping levels in AlN.

cond-mat.mtrl-sci↗

arXiv2026

PairUni: Pairwise Training for Unified Multimodal Language Models

Unified Vision-Language Models (UVLMs) perform both understanding and generation within a single architecture. Since these models rely on heterogeneous data and supervision, balancing both generation and understanding in reinforcement learning (RL) is challenging. To address this challenge, we propose PairUni, a unified framework that reorganizes data into understanding-generation (UG) pairs and aligns optimization accordingly. Specifically, we construct a unified paired dataset by synthesizing aligned instances via cross-modal semantic completion and retrieving semantically related samples. These paired structures expose cross-task semantic correspondences and support consistent policy learning. To leverage this structure, we present PairGRPO, a pair-aware variant based on Group Relative Policy Optimization. It assigns a similarity score to each pair to modulate the advantage, strengthening learning from well-aligned examples and reducing task interference. Extensive experiments across diverse UVLM architectures (Autoregressive and Discrete Diffusion) and scales (1B to 14B) demonstrate that PairUni yields consistent improvements over strong baselines. Notably, our method also demonstrates strong generalization by improving performance on image editing tasks without using any editing-specific data. Codes are available at https://github.com/Haochen-Wang409/PairUni.

cs.CL↗

arXiv2026

U-Net Based Image Enhancement for Short-time Muon Scattering Tomography

Muon Scattering Tomography (MST) is a promising non-invasive inspection technique, yet the practical application of short-time MST is hindered by poor image quality due to limited muon flux. To address this limitation, we propose a U-Net-based framework trained on Point of Closest Approach (PoCA) images reconstructed with simulation MST data to enhance image quality. When applied to experimental MST data, the framework significantly improves image quality, increasing the Structural Similarity Index Measure (SSIM) from 0.7232 to 0.9699 and decreasing the Learned Perceptual Image Patch Similarity (LPIPS) from 0.3604 to 0.0270. These results demonstrate that our method can effectively enhance low-statistics MST images, thereby paving the way for the practical deployment of short-time MST.

eess.IV↗

arXiv2026

Discovery of a 21 cm absorption system at z=2.327 with CHIME

We report the detection of a new 21 cm absorption system associated with the radio source NVSS J164725+375218 at a redshift of z=2.327, identified through a pilot survey conducted by the Canadian Hydrogen Intensity Mapping Experiment (CHIME). This is the fifth detection of an associated system at z > 2. By analyzing a subset of available data, we conduct a spectrally blind survey for 21 cm absorption systems within the redshift range of 0.78 to 2.55 along 202 lines of sight toward known sources in the declination range of 35 to 60 degrees. We detect three 21 cm absorbers: two previously known intervening systems and one newly discovered associated system. By fitting the absorption profiles with models containing one to three Gaussian components and selecting the best model using the Bayesian information criterion, we estimate the optical depth, velocity-integrated optical depth, and the ratio between the HI column density and the spin temperature of the absorption systems. These results demonstrate CHIME's ability to discover new absorbers, even in a small subset of its full dataset.

astro-ph.GA↗