https://techgazette.com.br/research/#vision This week's papers in Vision, each with what its authors did and found, in sentences from its own abstract. en 180 Thu, 15 Oct 2026 17:59:00 GMT https://arxiv.org/abs/2610.22371 https://arxiv.org/abs/2610.22371 Thu, 15 Oct 2026 17:59:00 GMT arXiv (vision, retrieval & HCI) cs.CV Vision arXiv (vision, retrieval & HCI) cs.CV — Did: We introduce Gen4P, a unified framework for recovering 4D geometry and motion from monocular video. Found: Together, these results demonstrate the potential of video generative pretraining for unified 4D perception. By Yitong Xia, Zeren Jiang, Chuanxia Zheng. https://arxiv.org/abs/2610.22370 https://arxiv.org/abs/2610.22370 Thu, 15 Oct 2026 17:59:00 GMT arXiv (vision, retrieval & HCI) cs.CV Vision arXiv (vision, retrieval & HCI) cs.CV — Did: In this work, we propose Dynamics-aware Token Merging (DaTMe), a training-free framework that grounds token merging in the underlying diffusion dynamics. Found: 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. By SeungJu Cha, Ye-Chan Kim, HyunGee Kim, Sungho Koh, Dong-Jin Kim. https://arxiv.org/abs/2610.22369 https://arxiv.org/abs/2610.22369 Thu, 15 Oct 2026 17:59:00 GMT arXiv (vision, retrieval & HCI) cs.CV Vision arXiv (vision, retrieval & HCI) cs.CV — Did: We propose SoftCache, which performs soft caching: instead of a binary switch, the headroom sets how far each cached step goes. Found: 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. By Shimon Malnick, Nadav Orenstein, Ohad Fried, Shai Avidan. https://arxiv.org/abs/2610.22363 https://arxiv.org/abs/2610.22363 Thu, 15 Oct 2026 17:59:00 GMT arXiv (vision, retrieval & HCI) cs.CV Vision arXiv (vision, retrieval & HCI) cs.CV — 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… Found: 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. By Zirui Wu, Chen Guo, Tianjian Jiang, Zeren Jiang, Zeming Li, Lingchen Yang. https://arxiv.org/abs/2610.22360 https://arxiv.org/abs/2610.22360 Thu, 15 Oct 2026 17:59:00 GMT arXiv (vision, retrieval & HCI) cs.CV Vision arXiv (vision, retrieval & HCI) cs.CV — 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. Found: In human-labeled sixteen-candidate pools, oracle selection improves complete factual success from 26.7% to 90.0%. By Shuanglin Li, Jiaying Lin, Shuquan Ye. https://arxiv.org/abs/2610.22356 https://arxiv.org/abs/2610.22356 Thu, 15 Oct 2026 17:58:00 GMT arXiv (vision, retrieval & HCI) cs.CV Vision arXiv (vision, retrieval & HCI) cs.CV — 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. Found: RC supervision further improves multi-step spatial reasoning, while the full model preserves general video understanding. By Han Su, Meng Tian, Tianyu Huang, Mingyu Gao, Lennox Wei, Dongxu Li. https://arxiv.org/abs/2610.22352 https://arxiv.org/abs/2610.22352 Thu, 15 Oct 2026 17:58:00 GMT arXiv (vision, retrieval & HCI) cs.CV Vision arXiv (vision, retrieval & HCI) cs.CV — 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. Found: 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. By Chen Yang, Jiemin Fang, Fanpeng Meng, Zihao Huang, Jichen Pang, Guanjun Wu. https://arxiv.org/abs/2610.22347 https://arxiv.org/abs/2610.22347 Thu, 15 Oct 2026 17:57:00 GMT arXiv (vision, retrieval & HCI) cs.CV Vision arXiv (vision, retrieval & HCI) cs.CV — Did: We introduce Speculative Forcing, a framework that treats intermediate clean estimates as drafts and predicts when individual tokens can be safely accepted early. Found: 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. By Hyunwook Choi, Seungryong Kim, Dongyoon Han, Sangdoo Yun. https://arxiv.org/abs/2610.22343 https://arxiv.org/abs/2610.22343 Thu, 15 Oct 2026 17:56:00 GMT arXiv cs.CV Vision arXiv cs.CV — Did: We propose GeoState, a view-decoupled geometry representation for predictive 3D world modeling that separates scene-state evolution from camera-dependent observation. Found: 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. By Hongchang Liu, Jian Liu, Yujuan Ding, Liang Peng, Xun Jiang, Fei Ma. https://arxiv.org/abs/2610.22341 https://arxiv.org/abs/2610.22341 Thu, 15 Oct 2026 17:56:00 GMT arXiv cs.CV Vision arXiv cs.CV — 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. Found: 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. By Yipeng Gao, Ziyao Zeng, Jinfa Huang, Yijiang Li, Haolin Xiong, Wang Qin. https://arxiv.org/abs/2610.22328 https://arxiv.org/abs/2610.22328 Thu, 15 Oct 2026 17:52:00 GMT arXiv cs.CV Vision arXiv cs.CV — Did: We introduce a training scheme that makes multi-turn image editing stable under this accumulating distribution shift. Found: Experiments show substantially reduced degradation over long multi-turn sequences while maintaining editing quality and instruction adherence. By Sara Dorfman, Nir Goren, Gal Barak, Daniel Cohen-Or, Or Patashnik. https://arxiv.org/abs/2610.22325 https://arxiv.org/abs/2610.22325 Thu, 15 Oct 2026 17:52:00 GMT arXiv (vision, retrieval & HCI) cs.CV Vision arXiv (vision, retrieval & HCI) cs.CV — 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. Found: Experiments on real people demonstrate substantially higher tracking accuracy and closer agreement with scanned surfaces than the evaluated baselines. By Rıza Alp Güler. https://arxiv.org/abs/2610.22321 https://arxiv.org/abs/2610.22321 Thu, 15 Oct 2026 17:51:00 GMT arXiv cs.CV Vision arXiv cs.CV — 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. Found: 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. By Yuruo Wu, Kaihang Pan, Kai Shen, Jiacheng Li, Shuyuan Tu, Jisheng Dang. https://arxiv.org/abs/2610.22320 https://arxiv.org/abs/2610.22320 Thu, 15 Oct 2026 17:51:00 GMT arXiv (vision, retrieval & HCI) cs.CV Vision arXiv (vision, retrieval & HCI) cs.CV — Did: We propose EOS-Story, a training-free framework that harmonizes identity consistency with scene-specific adaptation across text and vision spaces. Found: Extensive experiments show that EOS-Story achieves competitive or superior performance on both benchmarks. By SeungJu Cha, Ye-Chan Kim, Kwanyoung Lee, Dong-Jin Kim. https://arxiv.org/abs/2610.22311 https://arxiv.org/abs/2610.22311 Thu, 15 Oct 2026 17:50:00 GMT arXiv (vision, retrieval & HCI) cs.CV Vision arXiv (vision, retrieval & HCI) cs.CV — 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. Found: The fixed token-block granularity matches FlashAttention-style tiled execution and reduces the attention term that grows with rollout length. By Qiuhong Shen, Shizun Wang, Xinchao Wang. https://arxiv.org/abs/2610.22306 https://arxiv.org/abs/2610.22306 Thu, 15 Oct 2026 17:49:00 GMT arXiv (vision, retrieval & HCI) cs.CV Vision arXiv (vision, retrieval & HCI) cs.CV — Did: We present ARC-SORT (Adaptive, Robust, Compensated SORT), a tracking-by-detection tracker that trains nothing and runs in real time. By Girish Kumar Adari. https://arxiv.org/abs/2610.22295 https://arxiv.org/abs/2610.22295 Thu, 15 Oct 2026 17:44:00 GMT arXiv (vision, retrieval & HCI) cs.CV Vision arXiv (vision, retrieval & HCI) cs.CV — 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. Found: Finally, we demonstrate the limitations of existing evaluation metrics, which are overly reliant on human-annotations and exact text-text matching. By Jay N. Paranjape, Jurijs Nazarovs, Rohan Sarkar, Yue Guo, Lipeng Ke, Sankar Venkataraman. https://arxiv.org/abs/2610.22282 https://arxiv.org/abs/2610.22282 Thu, 15 Oct 2026 17:40:00 GMT arXiv (vision, retrieval & HCI) cs.CV Vision arXiv (vision, retrieval & HCI) cs.CV — Did: We instead ask whether stochastic exploration can be introduced inside the generator while preserving its original sampler. Found: 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. By Qiuhong Shen, Yiding Yang, Yuxin Zhang, Xinwei Huang, Minxuan Lin, Bo Liu. https://arxiv.org/abs/2610.22276 https://arxiv.org/abs/2610.22276 Thu, 15 Oct 2026 17:37:00 GMT arXiv (vision, retrieval & HCI) cs.CV Vision arXiv (vision, retrieval & HCI) cs.CV — 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. Found: 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… By Luozheng Qin, Yuhan Zhu, Changhai Zhou, Qian Qiao, Jun Gao, Yuhua Zhou. https://arxiv.org/abs/2610.22270 https://arxiv.org/abs/2610.22270 Thu, 15 Oct 2026 17:34:00 GMT arXiv (vision, retrieval & HCI) cs.CV Vision arXiv (vision, retrieval & HCI) cs.CV — 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. Found: These results show that color complements self-supervised features for route-level clustering, provided that it does not dominate the representation. By Juan Irving Vasquez. https://arxiv.org/abs/2610.22267 https://arxiv.org/abs/2610.22267 Thu, 15 Oct 2026 17:32:00 GMT arXiv (vision, retrieval & HCI) cs.CV Vision arXiv (vision, retrieval & HCI) cs.CV — 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. Found: 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. By Kwanyoung Lee, SeungJu Cha, Yebin Ahn, Sungho Koh, Dong-Jin Kim. https://arxiv.org/abs/2610.22264 https://arxiv.org/abs/2610.22264 Thu, 15 Oct 2026 17:30:00 GMT arXiv (vision, retrieval & HCI) cs.CV Vision arXiv (vision, retrieval & HCI) cs.CV — Did: We propose SCOPE, an objective based on Group Relative Policy Optimization (GRPO) that aligns update weighting with this structure. Found: 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. By Yufei Jiao, Jinfa Huang, Zeyue Xue, Tianyu Xie, Yuexiao Ma, Rongrong Ji. https://arxiv.org/abs/2610.22259 https://arxiv.org/abs/2610.22259 Thu, 15 Oct 2026 17:27:00 GMT arXiv (vision, retrieval & HCI) cs.CV Vision arXiv (vision, retrieval & HCI) cs.CV — 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. By James Willoughby, Irina Voiculescu. https://arxiv.org/abs/2610.22240 https://arxiv.org/abs/2610.22240 Thu, 15 Oct 2026 17:16:00 GMT arXiv (vision, retrieval & HCI) cs.CV Vision arXiv (vision, retrieval & HCI) cs.CV — Did: We introduce MMSearch-OPD, combining controllable data refinement with joint reinforcement learning (RL) and on-policy distillation (OPD). Found: Experiments show gains from data refinement and further improvements from joint training after SFT, indicating stronger long-horizon multimodal deep search capabilities. By Huanyao Zhang, Jiepeng Zhou, Jiaye Lin, Ontai Tang, Jingtian Hao, Mugeng Liu. https://arxiv.org/abs/2610.22232 https://arxiv.org/abs/2610.22232 Thu, 15 Oct 2026 17:12:00 GMT arXiv (vision, retrieval & HCI) cs.CV Vision arXiv (vision, retrieval & HCI) cs.CV — Did: Motivated by this insight, we propose WaveLLM, a wavelet-domain LLM-based entropy model for lossless image compression. Found: 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. By Yue Yu, Bowei Kang, Changsheng Gao, Dong Liu, Xiaoyan Sun.