arXiv subjects

Guilin Liu

Publications and source records attributed to Guilin Liu.

At least 37 records · Page 2Linked to original sources

arXiv2025

Argus: Vision-Centric Reasoning with Grounded Chain-of-Thought

Recent advances in multimodal large language models (MLLMs) have demonstrated remarkable capabilities in vision-language tasks, yet they often struggle with vision-centric scenarios where precise visual focus is needed for accurate reasoning. In this paper, we introduce Argus to address these limitations with a new visual attention grounding mechanism. Our approach employs object-centric grounding as visual chain-of-thought signals, enabling more effective goal-conditioned visual attention during multimodal reasoning tasks. Evaluations on diverse benchmarks demonstrate that Argus excels in both multimodal reasoning tasks and referring object grounding tasks. Extensive analysis further validates various design choices of Argus, and reveals the effectiveness of explicit language-guided visual region-of-interest engagement in MLLMs, highlighting the importance of advancing multimodal intelligence from a visual-centric perspective. Project page: https://yunzeman.github.io/argus/

cs.CV↗

arXiv2025

Nemotron-Research-Tool-N1: Exploring Tool-Using Language Models with Reinforced Reasoning

Enabling large language models with external tools has become a pivotal strategy for extending their functionality beyond text space. To enhance LLMs' tool-calling abilities, previous approaches primarily rely on supervised fine-tuning (SFT) with trajectories distilled from stronger models, often resulting in imitative reasoning that limits generalization. In this work, we explore rule-based reinforcement learning to enhance tool-calling in LLMs, resulting in Nemotron-Research-Tool-N1, a series of tool-calling reasoning models. Rather than enforcing supervision over intermediate distilled reasoning traces, Tool-N1 is trained with a binary RL reward that assesses only the format validity and functional correctness of tool invocations. This lightweight supervision allows the model to develop reasoning strategies independently, without relying on annotated trajectories. Experiments on several major benchmarks show that Tool-N1-7B/14B clearly outperform GPT-4o. We conduct a systematic study on the design of rule-based reinforcement learning strategies for training tool-calling models. Using 5,518 distilled reasoning trajectories, we compare SFT, RL, and the SFT-then-RL pipeline, finding that the widely adopted SFT-then-RL paradigm does not necessarily outperform pure RL.

cs.CL↗

arXiv2025

Eagle 2.5: Boosting Long-Context Post-Training for Frontier Vision-Language Models

We introduce Eagle 2.5, a family of frontier vision-language models (VLMs) for long-context multimodal learning. Our work addresses the challenges in long video comprehension and high-resolution image understanding, introducing a generalist framework for both tasks. The proposed training framework incorporates Automatic Degrade Sampling and Image Area Preservation, two techniques that preserve contextual integrity and visual details. The framework also includes numerous efficiency optimizations in the pipeline for long-context data training. Finally, we propose Eagle-Video-110K, a novel dataset that integrates both story-level and clip-level annotations, facilitating long-video understanding. Eagle 2.5 demonstrates substantial improvements on long-context multimodal benchmarks, providing a robust solution to the limitations of existing VLMs. Notably, our best model Eagle 2.5-8B achieves 72.4% on Video-MME with 512 input frames, matching the results of top-tier commercial model such as GPT-4o and large-scale open-source models like Qwen2.5-VL-72B and InternVL2.5-78B.

cs.CV↗

arXiv2025

Slow-Fast Architecture for Video Multi-Modal Large Language Models

Balancing temporal resolution and spatial detail under limited compute budget remains a key challenge for video-based multi-modal large language models (MLLMs). Existing methods typically compress video representations using predefined rules before feeding them into the LLM, resulting in irreversible information loss and often ignoring input instructions. To address this, we propose a novel slow-fast architecture that naturally circumvents this trade-off, enabling the use of more input frames while preserving spatial details. Inspired by how humans first skim a video before focusing on relevant parts, our slow-fast design employs a dual-token strategy: 1) "fast" visual tokens -- a compact set of compressed video features -- are fed into the LLM alongside text embeddings to provide a quick overview; 2) "slow" visual tokens -- uncompressed video features -- are cross-attended by text embeddings through specially designed hybrid decoder layers, enabling instruction-aware extraction of relevant visual details with linear complexity. We conduct systematic exploration to optimize both the overall architecture and key components. Experiments show that our model significantly outperforms self-attention-only baselines, extending the input capacity from 16 to 128 frames with just a 3% increase in computation, and achieving a 16% average performance improvement across five video understanding benchmarks. Our 7B model achieves state-of-the-art performance among models of similar size. Furthermore, our slow-fast architecture is a plug-and-play design that can be integrated into other video MLLMs to improve efficiency and scalability.

cs.CV↗

arXiv2025

GR00T N1: An Open Foundation Model for Generalist Humanoid Robots

General-purpose robots need a versatile body and an intelligent mind. Recent advancements in humanoid robots have shown great promise as a hardware platform for building generalist autonomy in the human world. A robot foundation model, trained on massive and diverse data sources, is essential for enabling the robots to reason about novel situations, robustly handle real-world variability, and rapidly learn new tasks. To this end, we introduce GR00T N1, an open foundation model for humanoid robots. GR00T N1 is a Vision-Language-Action (VLA) model with a dual-system architecture. The vision-language module (System 2) interprets the environment through vision and language instructions. The subsequent diffusion transformer module (System 1) generates fluid motor actions in real time. Both modules are tightly coupled and jointly trained end-to-end. We train GR00T N1 with a heterogeneous mixture of real-robot trajectories, human videos, and synthetically generated datasets. We show that our generalist robot model GR00T N1 outperforms the state-of-the-art imitation learning baselines on standard simulation benchmarks across multiple robot embodiments. Furthermore, we deploy our model on the Fourier GR-1 humanoid robot for language-conditioned bimanual manipulation tasks, achieving strong performance with high data efficiency.

cs.RO↗

arXiv2025

Eagle: Exploring The Design Space for Multimodal LLMs with Mixture of Encoders

The ability to accurately interpret complex visual information is a crucial topic of multimodal large language models (MLLMs). Recent work indicates that enhanced visual perception significantly reduces hallucinations and improves performance on resolution-sensitive tasks, such as optical character recognition and document analysis. A number of recent MLLMs achieve this goal using a mixture of vision encoders. Despite their success, there is a lack of systematic comparisons and detailed ablation studies addressing critical aspects, such as expert selection and the integration of multiple vision experts. This study provides an extensive exploration of the design space for MLLMs using a mixture of vision encoders and resolutions. Our findings reveal several underlying principles common to various existing strategies, leading to a streamlined yet effective design approach. We discover that simply concatenating visual tokens from a set of complementary vision encoders is as effective as more complex mixing architectures or strategies. We additionally introduce Pre-Alignment to bridge the gap between vision-focused encoders and language tokens, enhancing model coherence. The resulting family of MLLMs, Eagle, surpasses other leading open-source models on major MLLM benchmarks.

cs.CV↗

arXiv2025

Galactic-scale emission-line outflow from the radio-loud quasar 3C 191

Quasar feedback is routinely invoked as an indispensable ingredient in galaxy formation models. Galactic outflows are a crucial agent of quasar feedback that frequently manifest themselves in absorption and emission lines. Measuring the size and energetics of outflows based on absorption lines remains a challenge, and integral-field spectroscopy (IFS) mapping in emission lines is complementary. We present a VLT/SINFONI IFS mapping of quasar 3C 191 at $z \sim 2$, in which the outflow has been analyzed in absorption line spectroscopy. Three components are found based on the morphology and kinetics of [OIII]-emitting gas: a unshifted component which consistent with the systemic redshift and the location of the nucleus, a blueshifted in the north, and a redshifted in the south. The latter two components have velocities $\sim$ 600 km s$^{-1}$ and projected extents of 5 and 11 kpc, respectively, suggesting a biconical outflow structure. The blueshifted component's velocity is consistent with that derived from absorption lines. Using the electron density measured by the absorption lines and the luminosity and velocity of [OIII] outflow, we derive the mass outflow rate to be $\dot{M} \sim $ 9.5-13.4 M$_\odot$ yr$^{-1}$ and kinetic luminosity $\dot{E}_{\rm kin}$ ~ 2.5-3.7 $\times 10^{42}$ erg s$^{-1}$, consistent with absorption line analyses with VLT/Xshooter spectrum. The kinetic luminosity is only 0.01% of the bolometric luminosity, rendering a relatively weak outflow compared to typical expectation for effective feedback.

astro-ph.GA↗

arXiv2025

The distance of quasar outflows from the central source: The first consistent values from emission and absorption determinations

Measuring the distance of quasar outflows from the central source ($R$) is essential for determining their importance for AGN feedback. There are two methods to measure $R$: 1) A direct determination using spatially resolved Integral Field Spectroscopy (IFS) of the outflow in emission. 2) An indirect method which uses the absorption troughs from ionic excited states. The column density ratio between the excited and resonance states yields the outflow number density. Combined with a knowledge of the outflow's ionization parameter, $R$ can be determined. Generally, the IFS method probes $R$ range of several kpc or more, while the absorption method usually yields $R$ values of less than 1 kpc. There is no inconsistency between the two methods as the determinations come from different objects. Here we report the results of applying both methods to the same quasar outflow, where we derive consistent determinations of $R$ $\approx$ 5 kpc. This is the first time where the indirect absorption $R$ determination is verified by a direct spatially resolved IFS observation. In addition, the velocities (and energetics) from the IFS and absorption data are also found to be consistent. Therefore, these are two manifestations of the same outflow. In this paper we concentrate on the absorption $R$ determination for the outflow seen in quasar 3C 191 using VLT/X-shooter observations. We also reanalyze an older absorption determination for the outflow based on Keck/HIRES data and find that revised measurement to be consistent with ours. Our companion paper details the IFS analysis of the same object.

astro-ph.GA↗

arXiv2025

AIDE: Agentically Improve Visual Language Model with Domain Experts

The enhancement of Visual Language Models (VLMs) has traditionally relied on knowledge distillation from larger, more capable models. This dependence creates a fundamental bottleneck for improving state-of-the-art systems, particularly when no superior models exist. We introduce AIDE (Agentic Improvement through Domain Experts), a novel framework that enables VLMs to autonomously enhance their capabilities by leveraging specialized domain expert models. AIDE operates through a four-stage process: (1) identifying instances for refinement, (2) engaging domain experts for targeted analysis, (3) synthesizing expert outputs with existing data, and (4) integrating enhanced instances into the training pipeline. Experiments on multiple benchmarks, including MMMU, MME, MMBench, etc., demonstrate AIDE's ability to achieve notable performance gains without relying on larger VLMs nor human supervision. Our framework provides a scalable, resource-efficient approach to continuous VLM improvement, addressing critical limitations in current methodologies, particularly valuable when larger models are unavailable to access.

cs.CV↗

arXiv2025

Eagle 2: Building Post-Training Data Strategies from Scratch for Frontier Vision-Language Models

Recently, promising progress has been made by open-source vision-language models (VLMs) in bringing their capabilities closer to those of proprietary frontier models. However, most open-source models only publish their final model weights, leaving the critical details of data strategies and implementation largely opaque. In this work, we address VLM post-training from a data-centric perspective, showing the key role of data strategy in developing frontier VLMs. By studying and building our post-training data strategy from scratch, we share detailed insights into the development processes, aiming to benefit the development of competitive models for the open-source community. Our introduced data strategy, together with training recipes and model design, leads to a family of performant VLMs named Eagle2. Specifically, Eagle2-9B achieves state-of-the-art results across various multimodal benchmarks, matching certain competitive models with up to 70B parameters.

cs.CV↗

arXiv2025

Resolving turbulence drivers in two luminous obscured quasars with JWST/NIRSpec IFU

In this Letter, we investigate the turbulence and energy injection in the extended nebulae surrounding two luminous obscured quasars, WISEA J100211.29$+$013706.7 ($z=1.5933$) and SDSS J165202.64$+$172852.3 ($z=2.9489$). Utilizing high-resolution data from the NIRSpec IFU onboard the James Webb Space Telescope, we analyze the velocity fields of line-emitting gas in and around these quasars and construct the second-order velocity structure functions (VSFs) to quantify turbulent motions across different spatial scales. Our findings reveal a notable flattening in the VSFs from $\approx\!3$ kpc up to a scale of 10--20 kpc, suggesting that energy injection predominantly occurs at a scale $\lesssim$10 kpc, likely powered by quasar outflows and jet-driven bubbles. The extended spatial range of flat VSFs may also indicate the presence of multiple energy injection sources at these scales. For J1652, the turbulent energy in the host interstellar medium (ISM) is significantly higher than in tidally stripped gas, consistent with the expectation of active galactic nucleus (AGN) activities stirring up the host ISM. Compared to the VSFs observed on spatial scales of 10--50 kpc around lower-redshift UV-bright quasars, these obscured quasars exhibit higher turbulent energies in their immediate surroundings, implying different turbulence drivers between the ISM and halo-scale gas. Future studies with an expanded sample are essential to elucidate further the extent and the pivotal role of AGNs in shaping the gas kinematics of host galaxies and beyond.

astro-ph.GA↗

arXiv2024

CHANG-ES XXXV: Cosmic Ray Transport and Magnetic Field Structure of NGC 3556 at 3 GHz

Radio halos of edge-on galaxies are crucial for investigating cosmic ray propagation and magnetic field structures in galactic environments. We present VLA C-configuration S-band (2--4 GHz) observations of the spiral galaxy NGC 3556, a target from the Continuum Halos in Nearby Galaxies - an EVLA Survey (CHANG-ES). We estimate the thermal contribution to the radio emission from a combination of the H$α$ and mid-IR data, and employ Rotation Measure Synthesis to reveal the magnetic field structures. In our data, NGC 3556 exhibits a box-like radio halo extending nearly 7 kpc from the galactic plane. The scale height of the total S-band intensity in the halo is $1.68\pm 0.29$ kpc, while that of the non-thermal intensity is $1.93\pm 0.28$ kpc. Fitting the data to a 1-D cosmic-ray transport model, we find advection to describe the cosmic-ray propagation within the halo better than diffusion, with advection speeds of $245 \pm 15$ km s$^{-1}$ and $205 \pm 25$ km s$^{-1}$ above and below the disk, respectively. The magnetic field is detected patchily across the galaxy, displaying a toroidal configuration in the rotation measure map. The mean equipartition magnetic field strength is approximately $8.3\ μ$G in the disk and $4.5\ μ$G in the halo. In addition, a bubble-like structure extends nearly 3~kpc into the southern halo, aligned with the polarized intensity and H$α$ image, suggestive of superwinds generated by recent star formation feedback in the nuclear region.

astro-ph.GA↗

arXiv2024

First results from the JWST Early Release Science Program Q3D: The Fast Outflow in a Red Quasar at z=0.44

Quasar feedback may play a key role in the evolution of massive galaxies. The dust-reddened quasar, F2M110648.35$+$480712 at $z = 0.4352$ is one of the few cases at its redshift that exhibits powerful quasar feedback through bipolar outflows. Our new observation with the integral field unit mode of Near-infrared Spectrograph onboard JWST opens a new window to examine this spectacular outflow through Pa$α$ emission line with $\sim$3$\times$ better spatial resolution than previous work. The morphology and kinematics of the Pa$α$ nebula confirm the existence of a bipolar outflow extending on a scale of $\sim$17$\times$14 kpc and with a velocity reaching $\sim$1100 km s$^{-1}$. The higher spatial resolution of our new observation leads to more reliable measurements of outflow kinematics. Considering only the spatially resolved outflow and assuming an electron density of 100 cm$^{-2}$, the mass, momentum and kinetic energy outflow rates are $\sim$50-210 M$_{\odot}$ yr$^{-1}$, $\sim$0.3-1.7$\times$10$^{36}$ dynes ($\sim$14-78\% of the quasar photon momentum flux) and $\sim$0.16-1.27$\times$10$^{44}$ erg s$^{-1}$ ($\sim$0.02-0.20\% of the quasar bolometric luminosity), respectively. The local instantaneous outflow rates generally decrease radially. We infer that the quasar is powerful enough to drive the outflow, while stellar processes cannot be overlooked as a contributing energy source. The mass outflow rate is $\sim$0.4-1.5 times the star formation rate, and the ratio of kinetic energy outflow rate to the quasar bolometric luminosity is comparable to the minimum value required for negative quasar feedback in simulations. This outflow may help regulate the star formation activity within the system to some extent.

astro-ph.GA↗

arXiv2024

Could the inter-band lag of active galactic nucleus vary randomly?

The inter-band lags among the optical broad-band continua of active galactic nuclei (AGNs) have been intensively explored over the past decade. However, the nature of the lags remains under debate. Here utilizing two distinct scenarios for AGN variability, i.e., the thermal fluctuation of accretion disk and the reprocessing of both the accretion disk and clouds in the broad line region, we show that, owing to the random nature of AGN variability, the inter-band lags of an individual AGN would vary from one campaign with a finite baseline to another. Specifically, the thermal fluctuation scenario implies larger variations in the lags than the reprocessing scenario. Moreover, the former predicts a positive correlation between the lag and variation amplitude, while the latter does not result in such a correlation. For both scenarios, averaging the lags of an individual AGN measured with repeated and non-overlapping campaigns would give rise to a stable lag, which is larger for a longer baseline and gets saturation for a sufficiently long baseline. However, obtaining the stable lag for an individual AGN is very time-consuming. Alternatively, it can be equivalently inferred by averaging the lags of a sample of AGNs with similar physical properties, thus can be properly compared with predictions of AGN models. In addition, discussed are several new observational tests suggested by our simulations as well as the role of the deep high-cadence surveys of the Wide Field Survey Telescope in enriching our knowledge of the lags.

astro-ph.GA↗

arXiv2024

DiffiT: Diffusion Vision Transformers for Image Generation

Diffusion models with their powerful expressivity and high sample quality have achieved State-Of-The-Art (SOTA) performance in the generative domain. The pioneering Vision Transformer (ViT) has also demonstrated strong modeling capabilities and scalability, especially for recognition tasks. In this paper, we study the effectiveness of ViTs in diffusion-based generative learning and propose a new model denoted as Diffusion Vision Transformers (DiffiT). Specifically, we propose a methodology for finegrained control of the denoising process and introduce the Time-dependant Multihead Self Attention (TMSA) mechanism. DiffiT is surprisingly effective in generating high-fidelity images with significantly better parameter efficiency. We also propose latent and image space DiffiT models and show SOTA performance on a variety of class-conditional and unconditional synthesis tasks at different resolutions. The Latent DiffiT model achieves a new SOTA FID score of 1.73 on ImageNet256 dataset while having 19.85%, 16.88% less parameters than other Transformer-based diffusion models such as MDT and DiT,respectively. Code: https://github.com/NVlabs/DiffiT

cs.CV↗

arXiv2024

The Transition from Galaxy-wide Gas Inflow to Outflow in Quasar Host Galaxies

Galactic-wide outflows driven by active galactic nuclei (AGNs) is a routinely invoked feedback mechanism in galaxy evolution models. Hitherto, the interplay among the interstellar gas on galactic scales, the propagation of AGN outflows and the fundamental AGN parameters during evolution remains elusive. Powerful nuclear outflows are found to favorably exist at early AGN stages usually associated with high accretion rates and weak narrow emission lines. In a sample of quasars emitting Mg II narrow absorption lines (NALs) from the Sloan Digital Sky Survey, we discover an unprecedented phenomenon where galaxy-scale inflow-dominated transforming into outflow-dominated gas accompanied by an increasing strength of the narrow [O III] line, at a confidence level of 6.7σ. The fact that nuclear outflows diminish while galaxy-wide outflows intensifies as AGNs evolve implies that early-stage outflows interact with interstellar medium on galactic scales and trigger the gradual transformation into galaxy-wide outflows, providing observational links to the hypothetical multi-stage propagation of AGN outflows that globally regulates galaxy evolution.

astro-ph.GA↗

arXiv2024

USmorph: An Updated Framework of Automatic Classification of Galaxy Morphologies and Its Application to Galaxies in the COSMOS Field

Morphological classification conveys abundant information on the formation, evolution, and environment of galaxies. In this work, we refine the two-step galaxy morphological classification framework ({\tt\string USmorph}), which employs a combination of unsupervised machine learning (UML) and supervised machine learning (SML) techniques, along with a self-consistent and robust data preprocessing step. The updated method is applied to the galaxies with $I_{\rm mag}<25$ at $0.2<z<1.2$ in the COSMOS field. Based on their HST/ACS I-band images, we classify them into five distinct morphological types: spherical (SPH, 15,200), early-type disk (ETD, 17,369), late-type disk (LTD, 21,143), irregular disk (IRR, 28,965), and unclassified (UNC, 17,129). In addition, we have conducted both parametric and nonparametric morphological measurements. For galaxies with stellar masses exceeding $10^{9}M_{\sun}$, a gradual increase in effective radius from SPHs to IRRs is observed, accompanied by a decrease in the Sérsic index. Nonparametric morphologies reveal distinct distributions of galaxies across the $Gini-M_{20}$ and $C-A$ parameter spaces for different categories. Moreover, different categories exhibit significant dissimilarity in their $G_2$ and $Ψ$ distributions. We find morphology to be strongly correlated with redshift and stellar mass. The consistency of these classification results with expected correlations among multiple parameters underscores the validity and reliability of our classification method, rendering it a valuable tool for future studies.

astro-ph.GA↗

arXiv2024

JWST ERS Program Q3D: The pitfalls of virial BH mass constraints shown in a z = 3 quasar with an ultramassive host

We present JWST MIRI/NIRSpec observations of the extremely red quasar SDSS J165202.64+172852.3 at z~3, one of the most luminous quasars known to date, driving powerful outflows and hosting a clumpy starburst, amidst several interacting companions. We estimate the black hole (BH) mass of the system based on the broad H$α$ and H$β$ lines, as well as the Pa$β$ emission in the IR and MgII in the UV. We recover a very broad range of mass estimates, with constraints ranging between log $M_{\rm BH}$=9 and 10.1, which is exacerbated if imposing a uniform BLR geometry at all wavelengths. Several factors may contribute to the large spread: measurement uncertainties (insufficient sensitivity to detect the broadest component of the faint Pa$β$ line, spectral blending, ambiguities in the broad/narrow component distinction), lack of virial equilibrium (in a system characterised by powerful outflows and rapid accretion), and uncertainties on the luminosity-inferred size of the broad line region, a.o. given central dust obscuration. We constrain the stellar mass via SED fitting, suggesting the host to be extremely massive at $10^{12.8\pm 0.5} M_\odot$ - ~2 dex above the characteristic mass of the Schechter fit to the z=3 stellar mass function. Notably, J1652's central BH might be interpreted as being either undermassive, overmassive, or in line with the BH mass-stellar mass relation, depending on the choice of assumptions. The recovered Eddington ratio varies accordingly, but exceeds 10% in any case. We put our results into context by providing an extensive overview and discussion of recent literature results and their associated assumptions. Our findings provide an important demonstration of the uncertainties inherent in virial BH mass estimates, which are of particular relevance in the JWST era given the growing number of studies on rapidly accreting quasars at high redshift.

astro-ph.GA↗