arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2501.19084cs.CV
收藏

Laser:神经辐射场中高效的语言引导分割

Laser: Efficient Language-Guided Segmentation in Neural Radiance Fields

Durham University(杜伦大学);Institute of High Performance Computing (IHPC), ASTAR(高性能计算研究所(IHPC),新加坡科技研究局);School of Computer Science, China University of Geosciences(中国地质大学计算机学院);School of Computing, Newcastle University(纽卡斯尔大学计算机学院);University of Chinese Academy of Sciences(中国科学院大学)

Xingyu Miao, Haoran Duan, Yang Bai, Tejal Shah, Jun Song, Yang Long, Rajiv Ranjan, Ling Shao

AI总结:

本文提出Laser方法,利用CLIP特征蒸馏实现高效的语言引导3D场景分割,通过引入适配器模块、自交叉训练策略、低秩瞬态查询注意力机制及标签体分类策略,显著提升了分割精度、边缘准确性与颜色一致性,并在速度和性能上超越现有技术。

摘要:

在这项工作中,我们提出了一种利用CLIP特征蒸馏的方法,通过语言引导实现高效的3D分割。与以往依赖多尺度CLIP特征并受限于处理速度和存储要求的方法不同,我们的方法旨在通过直接且有效地蒸馏密集CLIP特征来简化工作流程,从而使用文本实现对3D场景的精确分割。为此,我们引入了适配器模块,并通过自交叉训练策略缓解了密集CLIP特征蒸馏过程中的噪声问题。此外,为了提高分割边缘的准确性,本工作提出了一种低秩瞬态查询注意力机制。为确保不同视点下相似颜色的分割一致性,我们通过标签体将分割任务转化为分类任务,这显著提升了颜色相似区域的分割一致性。我们还提出了一种简化的文本增强策略,以缓解CLIP特征与文本对应关系中的歧义问题。大量实验结果表明,我们的方法在训练速度和性能上均超越了当前最先进的技术。我们的代码可在以下网址获取:https://github.com/xingy038/Laser.git。

英文摘要:

In this work, we propose a method that leverages CLIP feature distillation, achieving efficient 3D segmentation through language guidance. Unlike previous methods that rely on multi-scale CLIP features and are limited by processing speed and storage requirements, our approach aims to streamline the workflow by directly and effectively distilling dense CLIP features, thereby achieving precise segmentation of 3D scenes using text. To achieve this, we introduce an adapter module and mitigate the noise issue in the dense CLIP feature distillation process through a self-cross-training strategy. Moreover, to enhance the accuracy of segmentation edges, this work presents a low-rank transient query attention mechanism. To ensure the consistency of segmentation for similar colors under different viewpoints, we convert the segmentation task into a classification task through label volume, which significantly improves the consistency of segmentation in color-similar areas. We also propose a simplified text augmentation strategy to alleviate the issue of ambiguity in the correspondence between CLIP features and text. Extensive experimental results show that our method surpasses current state-of-the-art technologies in both training speed and performance. Our code is available on: https://github.com/xingy038/Laser.git.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑