-
arXiv (vision, retrieval & HCI) cs.CV · Thursday, October 15 · Yitong Xia, Zeren Jiang, Chuanxia Zheng
- What they did
- We introduce Gen4P, a unified framework for recovering 4D geometry and motion from monocular video.
- What they found
Show
Together, these results demonstrate the potential of video generative pretraining for unified 4D perception.
The paper
-
arXiv (vision, retrieval & HCI) cs.CV · Thursday, October 15 · SeungJu Cha, Ye-Chan Kim, HyunGee Kim et al.
- What they did
- In this work, we propose Dynamics-aware Token Merging (DaTMe), a training-free framework that grounds token merging in the underlying diffusion dynamics.
- What they found
Show
Experiments on PixArt-a and FLUX demonstrate that DaTMe maintains visual fidelity across the entire image, highlighting diffusion dynamics as an effective token-level merging decision in efficient generation.
The paperCode
-
arXiv (vision, retrieval & HCI) cs.CV · Thursday, October 15 · Shimon Malnick, Nadav Orenstein, Ohad Fried et al.
- What they did
- We propose SoftCache, which performs soft caching: instead of a binary switch, the headroom sets how far each cached step goes.
- What they found
Show
SoftCache requires no training and adds no per-backbone settings: on FLUX for images and on HunyuanVideo and Wan2.1 for video, at speedups from $3.5\times$ to $6.9\times$, it is ahead of every baseline on all four fidelity metrics.
The paper
-
arXiv (vision, retrieval & HCI) cs.CV · Thursday, October 15 · Zirui Wu, Chen Guo, Tianjian Jiang et al.
- What they did
- We introduce P$^{3}$-WAM, a unified world action model for egocentric dexterous manipulation that learns 3D perception, action-conditioned prediction, and goal-directed planning through three complementary modes: Perceive, which reconstructs articulated hand motion and camera trajectories from video; Predict, which generates visual…
- What they found
Show
Through unified training, our model outperforms specialized baselines on 3D hand reconstruction and action-conditioned video prediction across four datasets, while producing goal-directed manipulation videos with stronger instruction adherence and action progression than video-generation baselines.
The paper
-
arXiv (vision, retrieval & HCI) cs.CV · Thursday, October 15 · Shuanglin Li, Jiaying Lin, Shuquan Ye
- What they did
- We argue that building reliable 3D systems requires not only stronger generation and reasoning, but also verification feedback that identifies unmet requirements and explains their violations, allowing systems to reflect on and refine their outputs.
- What they found
Show
In human-labeled sixteen-candidate pools, oracle selection improves complete factual success from 26.7% to 90.0%.
The paper
-
arXiv (vision, retrieval & HCI) cs.CV · Thursday, October 15 · Han Su, Meng Tian, Tianyu Huang et al.
- What they did
- We propose DK-VLM with Dual-Kernel Attention, which computes attention from both a visual token's input position and its geometric position in the scene.
- What they found
Show
RC supervision further improves multi-step spatial reasoning, while the full model preserves general video understanding.
The paper
-
arXiv (vision, retrieval & HCI) cs.CV · Thursday, October 15 · Chen Yang, Jiemin Fang, Fanpeng Meng et al.
- What they did
- To resolve these challenges, we introduce SUPRA-Tex, built on the principle of decoupling visible-material estimation, surface alignment, and observation-guided 3D completion.
- What they found
Show
Trained solely on public datasets, SUPRA-Tex achieves SOTA performance in both geometry-conditioned texturing and end-to-end asset generation, matching or even exceeding leading commercial systems on in-the-wild images.
The paper
-
arXiv (vision, retrieval & HCI) cs.CV · Thursday, October 15 · Hyunwook Choi, Seungryong Kim, Dongyoon Han et al.
- What they did
- We introduce Speculative Forcing, a framework that treats intermediate clean estimates as drafts and predicts when individual tokens can be safely accepted early.
- What they found
Show
On Lingbot-World 2.0, where camera motion continually changes the visible content, it maintains quality with ~2.5 denoising steps per token and runs 1.35x faster at 24.5 fps.
The paper
-
arXiv cs.CV · Thursday, October 15 · Hongchang Liu, Jian Liu, Yujuan Ding et al.
- What they did
- We propose GeoState, a view-decoupled geometry representation for predictive 3D world modeling that separates scene-state evolution from camera-dependent observation.
- What they found
Show
Experiments show that GeoState outperforms prior 3D world models on Waymo, supports geometrically consistent cross-view querying, and generalizes zero-shot to KITTI and nuScenes.
The paper
-
arXiv cs.CV · Thursday, October 15 · Yipeng Gao, Ziyao Zeng, Jinfa Huang et al.
- What they did
- We introduce 3DCodeVerse, a research ecosystem that enables coding agents to create editable and verifiable 3D content at scale. 3DCodeVerse comprises large-scale, high-quality datasets of executable 3D code, a coding harness for spatial reasoning and automated verification, and practical guidance for 3D coding.
- What they found
Show
Notably, our 3DCodeVerse harness achieves a 90% win rate against Claude Design (3D Object) in pairwise human evaluations, when both are utilized by Claude Opus 5.5.
The paper
-
arXiv cs.CV · Thursday, October 15 · Sara Dorfman, Nir Goren, Gal Barak et al.
- What they did
- We introduce a training scheme that makes multi-turn image editing stable under this accumulating distribution shift.
- What they found
Show
Experiments show substantially reduced degradation over long multi-turn sequences while maintaining editing quality and instruction adherence.
The paper
-
arXiv (vision, retrieval & HCI) cs.CV · Thursday, October 15 · Rıza Alp Güler
- What they did
- We introduce Dense4D, to our knowledge the first feed-forward model dedicated to dense 4D correspondence across the body, clothing and hair from a set of monocular RGB images, without a body template.
- What they found
Show
Experiments on real people demonstrate substantially higher tracking accuracy and closer agreement with scanned surfaces than the evaluated baselines.
The paper
-
arXiv cs.CV · Thursday, October 15 · Yuruo Wu, Kaihang Pan, Kai Shen et al.
- What they did
- We propose an event-centric evaluation framework that disentangles audio and visual events at the event level, analyzes each event independently, cross-validates relations across modalities, and aggregates the judgments into a coarse-to-fine, three-level consistency metric.
- What they found
Show
AVC-Reward achieves state-of-the-art alignment with human judgment among automatic methods, provides interpretable rationales that localize failures, and improves the audio-video consistency of a joint generation model when used as an online reinforcement-learning signal.
The paper
-
arXiv (vision, retrieval & HCI) cs.CV · Thursday, October 15 · SeungJu Cha, Ye-Chan Kim, Kwanyoung Lee et al.
- What they did
- We propose EOS-Story, a training-free framework that harmonizes identity consistency with scene-specific adaptation across text and vision spaces.
- What they found
Show
Extensive experiments show that EOS-Story achieves competitive or superior performance on both benchmarks.
The paper
-
arXiv (vision, retrieval & HCI) cs.CV · Thursday, October 15 · Qiuhong Shen, Shizun Wang, Xinchao Wang
- What they did
- To address this bottleneck, we introduce Velox, a training-free attention router for region-level context selection using frustum slab correspondences from streaming geometry.
- What they found
Show
The fixed token-block granularity matches FlashAttention-style tiled execution and reduces the attention term that grows with rollout length.
The paper
-
arXiv (vision, retrieval & HCI) cs.CV · Thursday, October 15 · Girish Kumar Adari
- What they did
- We present ARC-SORT (Adaptive, Robust, Compensated SORT), a tracking-by-detection tracker that trains nothing and runs in real time.
The paper
-
arXiv (vision, retrieval & HCI) cs.CV · Thursday, October 15 · Jay N. Paranjape, Jurijs Nazarovs, Rohan Sarkar et al.
- What they did
- In this work, we introduce the task of Grounded Human-Attributed Description and Activity Recognition (GHADAR), which involves describing the characteristics and activity descriptions of \textbf{every} person in a video, provided the location of the person in the video, which is more practical for real-world applications.
- What they found
Show
Finally, we demonstrate the limitations of existing evaluation metrics, which are overly reliant on human-annotations and exact text-text matching.
The paper
-
arXiv (vision, retrieval & HCI) cs.CV · Thursday, October 15 · Qiuhong Shen, Yiding Yang, Yuxin Zhang et al.
- What they did
- We instead ask whether stochastic exploration can be introduced inside the generator while preserving its original sampler.
- What they found
Show
Experiments on text-to-image and text-to-video generation show that RPO improves generation quality while achieving nearly 4x higher training efficiency than Flow-GRPO-style methods.
The paper
-
arXiv (vision, retrieval & HCI) cs.CV · Thursday, October 15 · Luozheng Qin, Yuhan Zhu, Changhai Zhou et al.
- What they did
- In light of this, we introduce MJev-Spatial, a System 1 multimodal decision model that tackles all three types of spatial understanding tasks within a unified classification framework.
- What they found
Show
Built on Qwen3.5-4B and trained on 12.3M question-answer pairs converted from existing spatial datasets, MJev-Spatial achieves the highest overall score (65.64%) among eight tested models, ranks first on seven of twelve benchmarks, and outperforms its unmodified autoregressive base model on all twelve, while running 2.5 times faster than…
The paper
-
arXiv (vision, retrieval & HCI) cs.CV · Thursday, October 15 · Juan Irving Vasquez
- What they did
- This study presents an unsupervised clustering methodology for public transport vehicles that combines morphological features extracted with a frozen DINOv2 Vision Transformer and handcrafted HSV color histograms through weighted early fusion, followed by UMAP projection and K-Means clustering.
- What they found
Show
These results show that color complements self-supervised features for route-level clustering, provided that it does not dominate the representation.
The paper
-
arXiv (vision, retrieval & HCI) cs.CV · Thursday, October 15 · Kwanyoung Lee, SeungJu Cha, Yebin Ahn et al.
- What they did
- We propose DetNO (Determinant Noise Optimization), which formulates diverse generation as RKHS volume maximization on the initial noise, maximizing the log-determinant of an RBF Gram matrix over one-step Tweedie predictions.
- What they found
Show
Fixing the pipeline and varying only the batch objective, both RBF-based aggregators outperform InfoNCE, and the log-determinant attains the best downstream accuracy on both benchmarks.
The paperCode
-
arXiv (vision, retrieval & HCI) cs.CV · Thursday, October 15 · Yufei Jiao, Jinfa Huang, Zeyue Xue et al.
- What they did
- We propose SCOPE, an objective based on Group Relative Policy Optimization (GRPO) that aligns update weighting with this structure.
- What they found
Show
Ablation studies show that both semantic boundaries and pooling within spans contribute to these gains, supporting semantic structure as a practical inductive bias for prompt engineering with RL.
The paper
-
arXiv (vision, retrieval & HCI) cs.CV · Thursday, October 15 · James Willoughby, Irina Voiculescu
- What they did
- We present FPG-Seg, an annotation-free pipeline for retinal vessel segmentation combining classical vesselness filtering, Bayesian deep learning and topology-aware graph reconstruction to segment retinal vessel topology without manual labels.
The paper
-
arXiv (vision, retrieval & HCI) cs.CV · Thursday, October 15 · Huanyao Zhang, Jiepeng Zhou, Jiaye Lin et al.
- What they did
- We introduce MMSearch-OPD, combining controllable data refinement with joint reinforcement learning (RL) and on-policy distillation (OPD).
- What they found
Show
Experiments show gains from data refinement and further improvements from joint training after SFT, indicating stronger long-horizon multimodal deep search capabilities.
The paper
-
arXiv (vision, retrieval & HCI) cs.CV · Thursday, October 15 · Yue Yu, Bowei Kang, Changsheng Gao et al.
- What they did
- Motivated by this insight, we propose WaveLLM, a wavelet-domain LLM-based entropy model for lossless image compression.
- What they found
Show
Built on Llama-3.2-3B with LoRA adaptation, WaveLLM achieves 2.51, 2.24, and 2.27 bpsp on Kodak, CLIC.pro, and DIV2K, respectively, outperforming the PixelCNN-based codec using the same wavelet transform by 0.43, 0.36, and 0.46 bpsp, and surpassing recent LLM-based lossless compressors on all three benchmarks.
The paper