Searcharxiv⌕ Search

arXiv subjects

Ruqi Huang

Publications and source records attributed to Ruqi Huang.

At least 19 recordsLinked to original sources
arXiv2026

SelfThinker: Self-Evolving Reasoning via Reliability-Verified Intrinsic Visual Thoughts

Unified multimodal models can both generate and understand images within a single set of weights, offering an opportunity to use image synthesis as training supervision for visual comprehension. However, existing visual chain-of-thought methods rely on human-annotated helper images during training and require slow image generation at test time. We present SelfThinker, the first self-evolving framework that turns a unified model's self-generated visual thoughts into stronger understanding without human image annotations. Given a visual problem, SelfThinker samples candidate helper images from its generation pathway and uses its frozen understanding pathway to score each candidate via conditional pointwise mutual information. High-utility visual thoughts then refine the generator via flow matching and serve as privileged teacher context for on-policy self-distillation. At inference, SelfThinker reasons directly from the original image in text without generating intermediate images. Although trained only on small synthetic visual-search and navigation datasets, SelfThinker transfers broadly to natural-image and spatial benchmarks, raising the seven-benchmark average of ThinkMorph-7B from 52.7 to 64.0 (+11.3 points). Controlled image-perturbation probes further show that SelfThinker surpasses ThinkMorph-7B by up to 40.4 points and its pretrained BAGEL-7B backbone by up to 23.2 points across fine-grained and spatial perception tasks. Our code is available at https://github.com/zhangquanchen/SelfThinker.

cs.CV↗
arXiv2026

Spatial-OPSD: Self-Improving Spatial Reasoning via Label-Free Self-Distillation

Vision-language models (VLMs) increasingly operate in embodied and spatially grounded settings, where accurate understanding of depth, viewpoint, and three-dimensional relations is essential. However, improving spatial reasoning typically relies on ground-truth answers, answer-derived rewards, or other forms of task-specific supervision. We introduce Spatial-OPSD, a label-free self-improvement framework that instead exploits spatial structure naturally available from perception and reconstruction tools. During training, a privileged teacher receives automatically obtainable spatial priors, such as depth, reconstructed 3D relations, and camera geometry, while the student observes only the original visual-language input. On trajectories sampled by the student itself, the teacher provides dense token-level supervision, allowing the student to internalize spatial knowledge without ground-truth answer labels or privileged information at inference time. To extend this supervision beyond a single round, we adopt a round-wise recursive training scheme: the teacher remains frozen within each round to provide a stable learning target, and the improved student initializes both teacher and student in the next round, where privileged spatial priors re-establish an informative teacher--student asymmetry. This enables repeated self-improvement while avoiding a rapidly moving teacher during optimization. Across four VLM families, a single round of Spatial-OPSD consistently improves the five-benchmark average, while three rounds further push a strong spatially specialized model to the open-source frontier, achieving the highest average among the open models and the best results on three of five spatial reasoning benchmarks. Our code is available at https://github.com/vermouth599/Spatial-OPSD.

cs.CV↗
arXiv2026

EviViT: Evidence-Adaptive Vision Transformers for Fine-Grained Perception

Fine-grained visual perception enables vision-language models to distinguish subtle attributes and ground their answers in visual evidence. In high-resolution scenes, processing the whole image at greater resolution spends visual tokens on irrelevant content, while isolated crops can lose the context needed to interpret the selected evidence. We introduce EviViT, a lightweight attachment that learns where a pretrained vision transformer should acquire detail. Human visual-search traces supervise a question-conditioned evidence density, which guides regional re-reading from the original pixels and the allocation of visual tokens. A sparse, coordinate-aware bridge then connects the regional features to the global scene, allowing the host to interpret precise evidence in context. Learned with the host backbone frozen, the attachment serves both the base model and compatible post-trained descendants without refitting. Experiments across nine hosts show consistent gains in average fine-grained accuracy. Matched-budget comparisons further show that EviViT outperforms global-only processing at every tested token ceiling while using fewer visual tokens.

cs.CV↗
arXiv2026

HaPRL: Human-Anchored Process Reinforcement Learning for Visual Search Agent

Multi-turn visual search agents answer questions about high-resolution images by iteratively deciding where to look. Reinforcement learning for these agents rewards only the final answer, leaving the search process unsupervised. Consequently, faulty routes in which the reasoning process is erroneous yet the final result is correct arise frequently, which in turn leads to ineffective training, i.e., scaling along the wrong paths. In this paper, we introduce HaPRL, the first framework to reinforce the search process with human search behavior. We first build an annotation platform and collect 1K+ human-annotated data with fine-grained behavioral signals. During training, a carefully designed judge scores each rollout with task-adaptive weights, anchored on the distilled trace of how a human annotator actually searched the same image. Extensive experiments show that HaPRL consistently outperforms outcome-based RL, and early-stage process supervision yields 6.7x more improvement in subsequent outcome-based scaling. Our results also demonstrate the importance of aligning model behavior with human process annotation signals, which offer new insight into the training of foundation models.

cs.CV↗
arXiv2026

SEGA: Drivable 3D Gaussian Head Avatar from a Single Image

Creating photorealistic 3D head avatars from limited input has become increasingly important for applications in virtual reality, telepresence, and digital entertainment. While recent advances like neural rendering and 3D Gaussian splatting have enabled high-quality digital human avatar creation and animation, most methods rely on multiple images or multi-view inputs, limiting their practicality for real-world use. In this paper, we propose SEGA, a novel approach for Single-imagE-based 3D drivable Gaussian head Avatar creation that combines generalized prior models with a new hierarchical UV-space Gaussian Splatting framework. SEGA seamlessly combines priors derived from large-scale 2D datasets with 3D priors learned from multi-view, multi-expression, and multi-ID data, achieving robust generalization to unseen identities while ensuring 3D consistency across novel viewpoints and expressions. We further present a hierarchical UV-space Gaussian Splatting framework that leverages FLAME-based structural priors and employs a dual-branch architecture to disentangle dynamic and static facial components effectively. The dynamic branch encodes expression-driven fine details, while the static branch focuses on expression-invariant regions, enabling efficient parameter inference and precomputation. This design maximizes the utility of limited 3D data and achieves real-time performance for animation and rendering. Additionally, SEGA performs person-specific fine-tuning to further enhance the fidelity and realism of the generated avatars. Experiments show our method outperforms state-of-the-art approaches in generalization ability, identity preservation, and expression realism, advancing one-shot avatar creation for practical applications.

cs.GR↗
arXiv2026

InceptionGS: Generative Bootstrapping for Large-Scale Gaussian Splatting under Unstructured View Sampling

Achieving truly immersive large-scale scene digitization necessitates consistent and visually pleasing rendering across all possible viewing perspectives. However, collecting multi-view images covering every fine detail of a large-scale scene is prohibitive due to scene complexity, capture cost, negligence, or accessibility constraints. As a result, the sampled views tend to be highly unstructured -- the majority of the scene is well covered yet certain regions inevitably lack sufficient observations. Existing reconstruction based methods are vulnerable to view scarcity while generation based approaches suffer from generalization, controllability, and 3D consistency issues. To address this challenge, we propose InceptionGS, which bootstraps Gaussian splatting by subtly balancing reconstruction and generation. Starting from an initial Gaussian splatting, InceptionGS reasonably rethinks and repairs problematic regions caused by view scarcity while preserving the quality elsewhere, by softly incorporating scene- and view-adaptive generative priors. Extensive experiments on real-world large-scale scenes demonstrate the superiority and broad applicability of our approach in handling unstructured imagery and boosting high-fidelity Gaussian splatting. Please refer to the supplementary video for better visual demonstrations.

cs.CV↗
arXiv2026

GS-Playground: A High-Throughput Photorealistic Simulator for Vision-Informed Robot Learning

Embodied AI research is undergoing a shift toward vision-centric perceptual paradigms. While massively parallel simulators have catalyzed breakthroughs in proprioception-based locomotion, their potential remains largely untapped for vision-informed tasks due to the prohibitive computational overhead of large-scale photorealistic rendering. Furthermore, the creation of simulation-ready 3D assets heavily relies on labor-intensive manual modeling, while the significant sim-to-real physical gap hinders the transfer of contact-rich manipulation policies. To address these bottlenecks, we propose GS-Playground, a multi-modal simulation framework designed to accelerate end-to-end perceptual learning. We develop a novel high-performance parallel physics engine, specifically designed to integrate with a batch 3D Gaussian Splatting (3DGS) rendering pipeline to ensure high-fidelity synchronization. Our system achieves a breakthrough throughput of 10^4 FPS at 640x480 resolution, significantly lowering the barrier for large-scale visual RL. Additionally, we introduce an automated Real2Sim workflow that reconstructs photorealistic, physically consistent, and memory-efficient environments, streamlining the generation of complex simulation-ready scenes. Extensive experiments on locomotion, navigation, and manipulation demonstrate that GS-Playground effectively bridges the perceptual and physical gaps across diverse embodied tasks. Project homepage: https://gsplayground.github.io.

cs.RO↗
arXiv2026

Video Models as Native 4D Renderers: World-Grounded Conditioning from Animated Mesh

Pretrained video diffusion models can act as renderers when the desired scene state is already specified by an animated mesh, a camera trajectory, and a reference image. This 4D generative rendering setting raises a representation question: what image-format condition lets a video backbone obey both camera motion and scene-internal animation? We propose DAR, a reference-guided renderer that extends Wan2.2 camera control from Plücker rays alone to a joint camera-plus-geometry interface. DAR projects a neural 4D G-buffer (tracking, world position, and normal) from the animated mesh and injects it through a widened control adapter while preserving the pretrained image-to-video prior. The central design choice is the pair of tracking and world position. Tracking identifies the persistent surface element that should carry appearance; world position gives its current scene-coordinate state; normal supplies local shape. Depth plus calibrated rays can recover 3D in principle, but depth is a camera-dependent chart in which camera and object motion are mixed. On the 68-case DAR-4D benchmark, LoRA DAR reaches PSNR 23.22, SSIM 0.895, and LPIPS 0.134, improving over off-the-shelf Wan2.2-Depth by 1.54 dB PSNR; a full fine-tune reaches PSNR 25.36 and SSIM 0.917. Matched ablations show that replacing world position by depth reduces PSNR by 1.26--1.55 dB at every checkpoint, supporting tracking+world-position correspondence as a practical 4D rendering condition.

cs.CV↗
arXiv2026

Agogic: Performance-Timed Music Tokens for LLM-Native Text-to-Symbolic-Music Generation

Text-to-music language models begin with a choice usually made by default: how to tokenize music. Normally entangled with backbone, data, and recipe, its effect has never been measured in isolation. We fix pretrained Qwen3.5 (0.8B-27B), data, budget, and decoding, and swap only the representation across seven tokenizations, anchoring texture metrics to each representation's model-free ceiling. The ordering is clean and surprising: representation, not model size, is the binding variable for distributional fidelity. Scaling the backbone 34x barely moves Frechet Music Distance (FMD), whereas switching representation halves it. PMT, a performance-resolution stream we release (10 ms timing, per-note velocity, multi-track texture; 609 symbols), reaches FMD 159 at 0.8B against 272-286 for beat grids (1.7-1.8x lower, up to 2.8x elsewhere; non-overlapping bootstrap CIs), so a 0.8B performance-resolution model beats a 27B beat grid. It reappears on a 26M from-scratch backbone and a second performance-resolution tokenizer: a property of the class, not one lucky vocabulary. Nor is it a finer-lattice artifact: snapping PMT's onsets to the beat grids' resolution still leaves it 67-129 FMD ahead of both (n=500). The effect is distributional; whether it is audible is a separate question, left open by our probe, with a human study pre-registered. Native caption adherence is weak but separable: a lightweight decode-time constraint doubles instrument-F1 (.28 to .60) and Correct-Key (.16 to .35) at no distributional cost. We release the harness, 25+ checkpoints, two corpora (86.6k aligned across caption/MIDI/ABC/audio; 6.25M captioned, the largest for music), and an imprinting diagnostic: published text-to-MIDI systems reproduce their training distribution near-invariant to the caption (72% vs. 71% chord-time on disjoint domains). The field's next representation claim can now be measured, not asserted.

cs.SD↗
arXiv2026

Bunraku: Turning a Single Illustration into an Editable Live2D Character

Live2D is the dominant 2D character-animation format for anime characters and virtual avatars, representing each character as a stack of RGBA layers driven by per-layer mesh deformation. Despite its wide use in virtual streaming, mobile games, and interactive characters, authoring a Live2D model still demands weeks of manual layer separation, occlusion completion, mesh placement, and keyframing, and no prior generative method produces such a structured asset end-to-end. We present the first system that, from a single illustration, generates all the structured information a Live2D runtime consumes: ordered RGBA layers, a deformation mesh per layer, and the parameter-driven keypose vertex offsets that make the character move. Stage 1 casts layered decomposition as a layered diffusion process under a Live2D-aware organ-level taxonomy, producing an ordered RGBA stack with hidden-region completion. Stage 2 builds a content-conforming triangle mesh for each layer from its alpha channel alone, then predicts the keypose displacement field of all layers jointly: every vertex of every layer is one token, self-attention spans layer boundaries, and each displacement is factorised into a bounded direction and a log-magnitude. Joint rather than independent prediction is what makes the result a coherent character instead of separately plausible parts, and is our largest gain; scaling the network 112x yields none. On 50 held-out characters, under true generation with no teacher forcing, Stage 2 attains a per-vertex direction cosine of 0.768 (median 0.828). Because a layer's mesh derives from its alpha channel, a clothing layer can be re-textured from a natural-language instruction while the mesh and predicted animation are reused byte-for-byte. We further contribute Live2D-Bench, the first standardized benchmark for the task, and an 8,884-model Live2D corpus with layer and animation supervision.

cs.CV↗
arXiv2026

Try Once, Then Optimal: De-Redundified Procedure Memory for Cross-Episode Exploration Amortization

Manipulating objects with hidden internal state, such as a latched microwave, forces a robot to probe before it can act. Yet a robot that has solved an instance once re-runs the same probes whenever it encounters that instance again, because existing cross-episode memories target task success and organize reuse around states, not the object or the cost of re-exploring it. We present Instance-Oriented Memory (IOM), an object-centric framework that amortizes this exploration: from a single encounter that uncovers the hidden state, whether or not it succeeds, IOM records a short procedure for manipulating that instance, keys it on the object's identifiable features, and injects it as a soft bias on a procedure-conditioned policy. A later encounter recognizes the object and recalls its procedure instead of re-exploring. We instantiate this distillation with an off-the-shelf vision-language model (VLM) that parses each encounter into the procedure without task-specific training. Across four articulated-object tasks, two in simulation (microwave, door) and two on a real robot (bottle, cabinet), an oracle procedure memory cuts manipulation operations by 16-30% over re-exploration at non-regressing success, and the VLM instantiation recovers 69-88% of that saving out of the box. Because the procedure is a soft bias on a feedback-driven policy, an incorrect memory is recovered from rather than obeyed: success holds even when a retrieved procedure is wrong, as for $\approx$12% of door instances. Across all tasks the benefit is purely one of efficiency: success never regresses, and on the real robot even improves. Code will be released upon acceptance.

cs.RO↗
arXiv2026

Engine-Native Editable 3D World Reconstruction with Objects and Lighting

Editable 3D scene creation requires object instances and lights that can be inspected, moved, and imported into standard engines, yet existing single-image methods largely stop at room-scale geometry, baked/global illumination, or text-driven generation. We introduce Lumera (Light-aware Unified Engine-native Reconstruction and Assembly), a benchmark and reference pipeline for engine-native, light-aware 3D scene parsing from a single image. Lumera-2K is built from 2,513 UE5 projects and provides 3.73M components, 63M object instances, 102.6K engine-native parametric lights, and 95.1K camera views. On this data, Lumera-Box and Lumera-Light adapt VLM to parse object boxes and parametric light tuples (x,y,z,r,g,b,I), which are assembled with per-object mesh reconstruction, HDR environment estimation, and a bounded agentic refinement loop. In a sanitized box benchmark against DetAny3D, SpatialLM, N3D-VLM, and WildDet3D, Lumera-Box obtains the strongest overall detection, geometry, semantic, and layout scores (merged mAP 0.1141, IoU-B 0.2472, F-score 0.2762), while WildDet3D remains stronger on anchor recall. For lights, Lumera-Light recovers almost all non-empty scenes (recall 0.998) but remains limited at individual-light localization (F1 0.209 at 0.5 m); matched lights have median position error 0.261 m, median ΔE2000 4.59, and intensity Pearson r=0.628. These results establish parametric lights as a measurable editable-scene target and expose remaining bottlenecks in relation structure, light recall/intensity, and cross-engine generalization.

cs.CV↗
arXiv2026

PairCoder++: Pair Programming as a Universal Paradigm for Verified Code-Driven Multimodal and Structured-Artifact Generation

Code is the medium through which large language models generate structured artifacts: charts, scientific figures, vector graphics, CAD models, 3D scenes, and hardware designs are all produced by writing programs. In this regime single pass inference is brittle, because the compiler, renderer, or simulator that decides whether the artifact exists is invisible to the model. We present PairCoder, which grounds review in the toolchain and realizes it as two agent pair programming: a Driver agent writes the program, a Navigator agent reviews it against verification evidence (diagnostics, execution results, and renderings of the current artifact beside the target), and the two switch roles when errors persist. Across 17 public benchmarks and seven models from three vendors, PairCoder improves essentially every benchmark whose artifact is verifiable, on full official metric suites rather than execution alone (for example, Blender scene executability 0.20 to 0.78; TikZ compile rate up 10 to 30 points on every model), at 2.9 to 9.2 times single model cost (about 7 times overall). The improvements concentrate where the toolchain provides an informative oracle and the baseline leaves headroom, and the method ties or mildly regresses where the oracle is weak; we frame pair programming as a reliable recipe for verified code driven generation.

cs.CL↗
arXiv2026

One Video, One World: Turning Monocular Video into Physical 4D Scenes

We introduce \textbf{OVOW}, the first training-free system that reconstructs \emph{instance-level, simulation-ready} 4D mesh scenes from a single monocular video. Recent 4D reconstruction achieves impressive rendering quality, but its outputs (\eg, implicit fields, Gaussian primitives, or point clouds) lack the watertight topology, instance separation, and standardized physical interfaces required by physics simulators and embodied AI. OVOW closes this gap with a four-stage pipeline: a vision-language model discovers, labels, and motion-classifies all instances; category-aware reconstruction yields per-instance meshes for rigid objects and topology-consistent mesh sequences for deformable ones; an iterative render-match-optimize procedure recovers metric scale and 6-DoF pose trajectories; and physics-grounded assembly enforces ground contact and inter-object support. Crucially, we model all motion, rigid and non-rigid, through direct vertex deformation without category-specific priors or skeleton rigging, producing watertight mesh scenes ready for downstream physics simulation and editing. We further establish the first benchmark for \emph{structured Video-to-4D} evaluation, with metrics for geometric correctness, instance separation, and physical plausibility beyond visual fidelity; the same pipeline doubles as a scalable engine for \emph{synthesizing} paired video-to-4D simulation data for future 4D world models and embodied AI. Across two synthetic benchmarks (static and 4D), OVOW attains the best overall layout and geometry accuracy and the lowest photometric and semantic error among all baselines, and on monocular video runs one to two orders of magnitude faster than the baselines, while downstream physics simulation confirms its physical stability.

cs.CV↗
arXiv2026

Exposure Bias Can Alleviate Itself via Directional and Frequency Rectification in Flow Matching

Flow Matching (FM) has achieved remarkable generative performance, yet it suffers from exposure bias due to discrepancies between training and inference. Existing mitigation strategies typically rely on static constraints or external heuristics. In this work, we propose that exposure bias itself inherently contains dynamic signals that can guide its own rectification. To leverage this, we introduce DEFAR (DirEctional-Frequency Adaptive Rectification). This framework simulates the single-step inference process during training to identify exposure bias. It utilizes directional and frequency-adaptive feedback signals from the bias itself to enhance the model's bias tolerance. It consists of two key components: (1) Anti-Drift Rectification (ADR). ADR treats inference-time drift as a signal to learn the direction to steer deviated states back toward the target. ADR endows the model with intrinsic active self-rectification capabilities; (2) Frequency Compensation (FC). Empirically, we observe that accumulated bias often stems from a lack of low-frequency components in high-noise stages, and exposure bias carries the missing frequency. FC leverages the bias itself as a self-feedback weighting factor to reinforce the missing frequency components. Experiments on CIFAR-10, CelebA-64, and ImageNet-256/512 show that DEFAR outperforms prior baselines and further demonstrates favorable scalability, compatibility, and inference robustness.

cs.CV↗
arXiv2026

DANTE-W: Diffuse Albedo Neural Texturing in the Wild

Classical mesh texturing techniques blend captured multi-view images directly, which inevitably suffer from baked-in shading and casted shadows that compromise visual fidelity during relighting. To circumvent this issue, we present a neural texturing framework, namely DANTE-W, to enable high-fidelity diffuse albedo texture recovery from unstructured image collections for large-scale, in-the-wild scenes, which integrates seamlessly with traditional 3D reconstruction pipelines. Given a reconstructed mesh and its surface parameterization, our method fuses view-space generative albedo priors into a coherent texture space via an expressive neural representation, while substantially enhancing fine-grained textural details through physically principled neural rendering. To comprehensively evaluate our method, we curate a benchmark dataset featuring diverse, fine-grained textures, comprising both real-world in-the-wild scenes and synthetic objects. Extensive experiments verify the effectiveness of our approach in reconstructing accurate albedo textures and boosting relighting fidelity. Project page: dante-wild.github.io.

cs.GR↗
arXiv2026

FILIC: Dual-Loop Force-Guided Imitation Learning with Impedance Torque Control for Contact-Rich Manipulation Tasks

Many contact-rich manipulation tasks require precise force regulation. However, most imitation learning (IL) policies remain position-centric and lack explicit force awareness, and adding force/torque sensors to collaborative robot arms is often costly and requires additional hardware design. To overcome these issues, we propose FILIC, a Force-guided Imitation Learning framework with impedance torque control. FILIC integrates a Transformer-based IL policy with an impedance controller in a dual-loop structure, enabling compliant force-informed, force-executed manipulation. For robots without force/torque sensors, we introduce a cost-effective end-effector force estimator using joint torque measurements through analytical Jacobian-based inversion while compensating with model-predicted torques from a digital twin. Experiments show that FILIC significantly outperforms vision-only and joint-torque-based methods, achieving safer, more compliant, and adaptable contact-rich manipulation. The source code is available at https://github.com/OpenGHz/mujoco-wrench-estimator.git.

cs.RO↗
arXiv2026

When Video Misreads: Closed-Loop Distillation of Reading Heuristics for Exploratory Manipulation Trace QA

Exploratory manipulation often turns an apparent failed attempt into the key evidence for what to do next. For example, a robot pulls a locked cabinet drawer, fails, and only succeeds after opening the lock. The failed pull reveals a latent precondition (the drawer is locked) that determines the minimal-success action chain (the fewest actions that complete the task), here [lock-open, drawer-pull]. Correctly reading this trace is therefore the prerequisite for recovering that chain. We formalize this setting as Exploratory Manipulation Trace QA (EMT-QA): given synchronized video and proprioception from an exploratory trace, predict the minimal-success action chain under the latent precondition revealed by the probe. However, even state-of-the-art VLMs and embodied multimodal LLMs misread this evidence: they do not reliably recover the chain from raw video, raw proprioception, or their combination. We introduce Closed-Loop Trace Distillation, a pipeline that uses a per-task coding agent to inspect labeled training traces and distill a one-line natural-language prompt over the trace, which we call the Distilled Reading Heuristic (DRH). At inference, no agent is invoked and no model weights are updated; a frozen VLM receives the raw trace plus the DRH as a prompt entry. Across three simulator and two real-robot tasks, the DRH improves chain accuracy by +0.38 to +0.47 over the best raw-modality baseline. The same DRH also serves as the sole specification for one-shot programmatic classifiers that match the prompted VLM.

cs.RO↗