arXiv Computer Vision

Beyond Argmax: A Mechanistic Study of Semantic Retention in Frozen Foundation-Model Composition for Generalized Few-Shot 3D Segmentation

The paper investigates how much semantic information is lost when frozen foundation models are combined for few‑shot 3D segmentation. By varying the number of retained semantic alternatives before fusion, the authors show that keeping the full distribution of class scores yields higher harmonic‑mean IoU than collapsing to a single class. Experiments on ScanNet200 and ScanNet++ confirm that full‑distribution fusion consistently outperforms top‑1 and other operators, and that most useful information is recovered by retaining a compact set of plausible alternatives.

arXiv Machine Learning
Aug 28

Generative Semantic Scene Completion

The paper introduces Generative Semantic Scene Completion (GSSC), a framework that recasts outdoor LiDAR semantic scene completion as a discrete diffusion process. It comprises three components: (1) paired sparse‑dense scene synthesis (PS³) to generate synthetic training data, (2) semantic‑guided generative scene completion (SGSC) that generates scenes from noise conditioned on sparse scans, and (3) structured source discrete diffusion (S²D²) that refines existing completions in a single flow‑matching step. Using this approach, the authors achieve state‑of‑the‑art performance on the SemanticKITTI benchmark, reaching 38.8% mIoU in a single‑sweep, single‑sample setting and 39.2% with limited augmentation.

By Shi Chen, Weifeng Ge
arXiv Computer Vision
5d ago

Retrieved Images as Visual Thought: Training-Free Multimodal In-Context Learning for the Open-vs-Closed Gap

ReVisIT is a train‑free framework that turns retrieved image‑label pairs into units of visual thought, combining structured class definitions, multimodal retrieval, and alternating user/assistant injection before joint decoding. On several benchmarks—including Fast Open MiniImageNet, Bongard‑OpenWorld, and the newly released MAAC‑Bench—ReVisIT achieves performance comparable to or surpassing large, trained models while using far fewer parameters. The approach demonstrates that high‑quality retrieval and a simple turns layer can provide a universal performance boost across diverse multimodal tasks.

By Bingchen Huang, Zhiling Wang, Yifu Chen, Yuanchao Du
arXiv AI
Sep 2

VOIM: Training-Free Open-Vocabulary 3D Instance Mapping for RGB-D and Monocular SLAM

VOIM (Voxel‑Grounded Online Instance Manager) is a training‑free system that builds open‑vocabulary 3D instance maps from RGB‑D or monocular RGB input by deferring label and instance decisions until sufficient soft evidence accumulates per voxel across views. Across four perception configurations on ScanNet++, VOIM outperforms the strongest online RGB‑D system, OVO‑SLAM, by 4.8–11.7 mIoU, and achieves 44.07 mIoU under a like‑for‑like protocol, winning all ten scenes. The method also runs unchanged on monocular RGB, matching baseline performance on Replica, and produces exportable occupancy grids that support free‑form instance queries.

By Sangmin Song, Sarath Kodagoda, Marc G. Carmichael, Karthick Thiyagarajan, Amal Gunatilake, Kelly Prentice, Jodi Martin
arXiv Computer Vision
Aug 28

When Semantics Saturate or Emerge: Adaptation-Conditional Semantic Utility in Source-Free Cross-Domain Few-Shot Learning

The paper investigates whether language prompts selected by zero‑shot accuracy remain effective after visual adaptation in source‑free cross‑domain few‑shot learning. Using a paired protocol, the authors compare generic class‑name templates with detailed class descriptions before and after Low‑Rank Adaptation (LoRA) on datasets such as EuroSAT, CropDisease, ISIC, and ChestX. They identify two regimes: semantic saturation, where detailed prompts are already useful before adaptation, and semantic emergence, where detailed prompts become more useful only after visual representation updates, driven by changes in prediction patterns.

By Wei Liu, Xing Deng, Haijian Shao
arXiv AI
Aug 19

Where a New Concept Must Enter: Entry Point Gates Cross-Task Usability in Unified Multimodal Models

The paper investigates how new concepts can be integrated into unified multimodal models (UMMs) by separating generation and understanding objectives through a novel visual entity bound to a single task direction. Experiments show that the effectiveness of cross‑task usability depends on where the concept is injected into the shared computation, with a mid‑stack alignment objective achieving high concept acquisition with minimal loss to overall performance. The study highlights that unified weights alone are insufficient; the two directions must share a semantic format at the entry point for efficient concept integration.

By Zongyang Qiu, Yihan Wu, Kaixuan Fan, Bo Li, Hui Xiong
arXiv AI
Sep 1

When Images Look Right and Retrieve Wrong: Coverage-Guided Cross-Scale Re-Indexing for Knowledge-Faithful Generative Perception

The paper introduces CERES, a closed‑loop multimodal indexing framework that addresses semantic collapse in multimodal generation by building a three‑level semantic pyramid and using scale‑routed cross‑attention to generate images that remain retrievable by their original queries. CERES employs a co‑occurrence‑aware router, a lightweight U‑Net generator, and a soft‑Jaccard coverage objective to ensure generated images cover the intended concepts, verified by re‑indexing with a frozen vision‑language model and an external DINOv2 probe. Experiments on four pansharpening benchmarks show state‑of‑the‑art performance, especially under extreme scale variation, and significant improvements in concept‑query retrieval and image‑text ranking metrics.

By Guangyuan Dong, Chuang Liu, Haoyu Wang, Yangchen Zeng, Jiaqi Zhang, Li Jiuxing, Xiaoyang Yu, Pinlong Zhao, Yuchao Hou, Ziwei Li, Zheng Lin, Alexander Lim Han Yang, Yusen Wu