arXivDaily arXiv每日学术速递 周一至周五更新
​通知:联邦节假日调休:哥伦布日 / 原住民日(Columbus Day / Indigenous Peoples' Day),今日(13日)无更新。
arXiv 2310.14108cs.LGcs.AIcs.CV

CLIP 遇见 Model Zoo 专家:用于视觉增强的伪监督

CLIP meets Model Zoo Experts: Pseudo-Supervision for Visual Enhancement

University of Washington(华盛顿大学);Apple, Inc.(苹果公司)

Mohammadreza Salehi, Mehrdad Farajtabar, Maxwell Horton, Fartash Faghri, Hadi Pouransari, Raviteja Vemulapalli, Oncel Tuzel, Ali Farhadi, Mohammad Rastegari, Sachin Mehta

AI总结:

本文提出利用模型动物园中的任务特定视觉模型生成伪标签,辅助 CLIP 进行对比训练,在不损害其零样本分类能力的前提下,显著提升了分割、检测等视觉任务的性能。

摘要:

对比语言图像预训练(CLIP)是训练视觉语言模型的标准方法。尽管 CLIP 具有可扩展性、可提示性,且在图像分类任务中对分布偏移具有鲁棒性,但它缺乏目标定位能力。本文研究了以下问题:我们能否利用模型动物园中的任务特定视觉模型来增强 CLIP 训练,以改善其视觉表示?为此,我们利用开源的任务特定视觉模型,为未经整理的含噪图像文本数据集生成伪标签。随后,除了对图像和文本对进行对比训练外,我们还基于这些伪标签训练 CLIP 模型。这一简单设置在包括分割、检测、深度估计和表面法线估计在内的不同视觉任务中,展现出高达 16.3% 的显著提升。重要的是,这些增强是在不损害 CLIP 现有能力(包括其在可提示零样本分类中的熟练度)的情况下实现的。

英文摘要:

Contrastive language image pretraining (CLIP) is a standard method for training vision-language models. While CLIP is scalable, promptable, and robust to distribution shifts on image classification tasks, it lacks object localization capabilities. This paper studies the following question: Can we augment CLIP training with task-specific vision models from model zoos to improve its visual representations? Towards this end, we leverage open-source task-specific vision models to generate pseudo-labels for an uncurated and noisy image-text dataset. Subsequently, we train CLIP models on these pseudo-labels in addition to the contrastive training on image and text pairs. This simple setup shows substantial improvements of up to 16.3% across different vision tasks, including segmentation, detection, depth estimation, and surface normal estimation. Importantly, these enhancements are achieved without compromising CLIP's existing capabilities, including its proficiency in promptable zero-shot classification.

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。

↑