arXiv subjects

Wenfeng Deng

Publications and source records attributed to Wenfeng Deng.

6 recordsLinked to original sources

arXiv2026

CA-World: Multi-Object Counterfactual Alignment for Efficient Interactive-Ready Reconstruction

Reconstructing interaction-ready 3D worlds is essential for physical simulation, virtual reality, robotics, and autonomous driving. However, existing methods mainly optimize static and holistic visual fidelity, with limited support for multi-object interaction. We argue that an interaction-ready reconstruction should anticipate potential scene changes and preserve geometric completeness, visual quality, multi-object spatial relationship, and physical plausibility under potential interactions. To this end, motivated by the causal intervention, we propose CA-World, an efficient framework that integrates counterfactual alignment learning into a decoupling-reintegration reconstruction pipeline. Specifically, we formulate foreground-background decoupling as a visual intervention, separate object generation and background inpainting as counterfactual generation, and scene reintegration as an inverse intervention. According to counterfactual consistency, reversing the intervention should recover the factual world, motivating three alignment objectives between the reintegrated and original scenes: appearance, spatial, and physical consistency. This formulates interaction-ready reconstruction as counterfactual alignment learning with direct supervision. Moreover, leveraging the locality of object-level interventions, CA-World constrains counterfactual states using the observed scene, enabling efficient and coherent reintegration without jointly optimizing all object states, thereby reducing computational cost and error accumulation. Experiments on object completeness, spatial accuracy, outdoor background completion, rendering quality, simulated dynamics, and downstream applications demonstrate the effectiveness of CA-World. Project page: https://chnxindong.github.io/ca-world/.

cs.CV↗

arXiv2026

Occlusion-Robust Multi-Object Decoupling for Physics-Based Robotic Interaction

We propose a mask-free method for lossless multi-object 3D reconstruction from sparse and occluded real-world views, enabling physically plausible robotic interaction via Material Point Method (MPM) simulation. Our key insight is that object coupling stems from occlusion and limited viewpoints, which we address by formulating multi-object decoupling as a sparse-view reconstruction problem. Using 3D Gaussian Splatting as base representation, we first obtain coarse instance partitions with a SAM2-trained segmentation field. Rather than relying on masks, we reconstruct fragmented geometries by leveraging a joint Score Distillation Sampling (SDS) process, which integrates reference-view supervision with novel-view synthesis guided by 2D and 3D diffusion priors to enforce both texture fidelity and 3D consistency. Furthermore, we incorporate geometry-aware priors such as intra-object and inter-object similarity to regularize geometric reasoning. Experimental results demonstrate that our method produces complete, simulation-ready 3D objects without requiring manual masks, enabling realistic dynamic interactions on both synthetic, robotic and real-world datasets.

cs.CV↗

arXiv2026

STARE: Surprisal-Guided Token-Level Advantage Reweighting for Policy Entropy Stability

Reinforcement Learning with Verifiable Rewards algorithms like GRPO have emerged as the dominant post-training paradigm for complex reasoning in LLMs, yet commonly suffer from policy entropy collapse during training. We conduct a first-order gradient analysis of token-level entropy dynamics under GRPO and identify a token-level credit assignment mismatch: the per-token entropy variation decomposes into the product of the trajectory-level advantage and an entropy sensitivity function over the next-token distribution, yielding an advantage-surprisal four-quadrant structure and a near-criticality property. Motivated by it, we propose STARE (Surprisal-guided Token-level Advantage Reweighting for policy Entropy stability), which identifies entropy-critical token subsets via batch-internal surprisal quantiles, selectively reweights their effective advantages, and incorporates a target-entropy closed-loop gate for stable entropy regulation. Across model scales from 1.5B to 32B and three task families (Short CoT, Long CoT, and Multi-Turn Tool Use), STARE sustains stable RL training over thousands of steps while maintaining policy entropy within the target band. On AIME24 and AIME25, STARE outperforms DAPO and other competitive baselines by 4%-8% in average accuracy, with reflection tokens and response length growing in tandem, indicating sustained exploration-exploitation balance that further unlocks RL training potential.Code is available at https://github.com/hp-luo/STARE.

cs.LG↗

arXiv2026

CoSTL: Comprehensive Spatial-Temporal Representation Learning for Moment Retrieval and Highlight Detection

Video Moment Retrieval (MR) and Highlight Detection (HD) are crucial tasks in video analysis that aim to localize specific moments and estimate clip-wise relevance based on a given text query. Recent approaches treat them as similar video grounding tasks and use the same architecture to solve them. These tasks require both fine-grained comprehension at the image level and high-level temporal understanding across the entire video. Existing approaches have primarily focused on temporal modeling using frame-level features, often neglecting the rich visual information related to the text query within individual frames. This oversight leads to inaccurate grounding results. To address this limitation, we propose a Comprehensive Spatial-Temporal Representation Learning Framework (CoSTL), which captures both fine-grained image-level information and temporal dynamics. Specifically, CoSTL incorporates a text-driven progressive fine-grained image encoder, performing a two-step text-driven knowledge extraction process to learn fine-grained spatial representations. Furthermore, a multi-scale temporal perception module captures comprehensive spatial-temporal representations, enhancing the model's ability to process temporal dynamics. We demonstrate state-of-the-art performance on four public benchmarks: QVHighlights, Charades-STA, TACoS, and TVSum.

cs.CV↗

arXiv2026

Boosting Zero-Shot 3D Style Transfer with 2D Pre-trained Priors

In this work, we focus on zero-shot 3D style transfer that can generate multi-view consistent stylized views of the 3D scene given an arbitrary style image. We primarily tackle the issue of data scarcity in 3D style transfer, which arises when each model is trained on only a single scene, thereby limiting the number of available content images. This scarcity significantly hampers stylization performance, as model optimization relies on a sufficient number of content-style image pairs to provide supervisory signals. Our core idea is to integrate a decoder pre-trained on large-scale 2D image datasets into the 3D style transfer pipeline, thereby leveraging the prior knowledge encoded in the decoder from learning over numerous content-style image pairs. Our method combines feature Gaussian splatting and deferred stylization, enabling high-quality stylization with the data-sufficient decoder network while ensuring view consistency by unifying view-dependent operations into a view-invariant process. Experiments demonstrate that our Data-Sufficient StyleGaussian (DS-StyleGaussian) model outperforms existing zero-shot 3D style transfer methods in terms of visual quality across various datasets. This work also suggests that 2D pre-training can serve as a strong enhancement for 3D tasks, bridging the data gap between 2D and 3D.

cs.CV↗

arXiv2025

Multi-Satellite Beam Hopping and Power Allocation Using Deep Reinforcement Learning

In non-geostationary orbit (NGSO) satellite communication systems, effectively utilizing beam hopping (BH) technology is crucial for addressing uneven traffic demands. However, optimizing beam scheduling and resource allocation in multi-NGSO BH scenarios remains a significant challenge. This paper proposes a multi-NGSO BH algorithm based on deep reinforcement learning (DRL) to optimize beam illumination patterns and power allocation. By leveraging three degrees of freedom (i.e., time, space, and power), the algorithm aims to optimize the long-term throughput and the long-term cumulative average delay (LTCAD). The solution is based on proximal policy optimization (PPO) with a hybrid action space combining discrete and continuous actions. Using two policy networks with a shared base layer, the proposed algorithm jointly optimizes beam scheduling and power allocation. One network selects beam illumination patterns in the discrete action space, while the other manages power allocation in the continuous space. Simulation results show that the proposed algorithm significantly reduces LTCAD while maintaining high throughput in time-varying traffic scenarios. Compared to the four benchmark methods, it improves network throughput by up to $8.9\%$ and reduces LTCAD by up to $69.2\%$

eess.SY↗