arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2304.08014cs.CV
收藏

使用几何变换敏感架构从非以物体为中心的图像中进行自监督学习

Self-Supervised Learning from Non-Object Centric Images with a Geometric Transformation Sensitive Architecture

Hanyang University(汉阳大学)

Taeho Kim, Jong-Min Lee

AI总结:

针对非以物体为中心图像中因裁剪改变语义及模型难以捕获位置信息的问题,提出几何变换敏感架构(GTSA),通过预测旋转、池化旋转教师特征图及图像块对应损失使模型对几何变换敏感,在多项视觉任务上显著超越 DINO 基线。

摘要:

大多数基于不变性的自监督方法依赖于单张以物体为中心的图像(例如 ImageNet 图像)进行预训练,学习对几何变换不变的特征。然而,当图像不是以物体为中心时,由于裁剪,图像的语义可能会发生显著改变。此外,随着模型对几何变换变得不敏感,它可能难以捕获位置信息。出于这个原因,我们提出了一种几何变换敏感架构(Geometric Transformation Sensitive Architecture),旨在对几何变换敏感,特别关注四折旋转、随机裁剪和多裁剪。我们的方法通过预测旋转并使用通过池化和旋转教师特征图而随这些变换变化的目标,鼓励学生模型变得敏感。此外,我们使用图像块对应损失来鼓励具有相似特征的图像块之间的对应关系。这种方法使我们能够以更合适的方式捕获长程依赖,而不是通过鼓励局部到全局的对应关系(这在学习对多裁剪不敏感时发生)来捕获长程依赖。与训练模型对几何变换不敏感的其他方法相比,当使用非以物体为中心的图像作为预训练数据时,我们的方法展现出了提升的性能。我们在包括图像分类、语义分割、目标检测和实例分割在内的任务中超越了 DINO [Caron et al. [2021b]] 基线,提升幅度分别为 4.9 $Top-1 Acc$、3.3 $mIoU$、3.4 $AP^b$ 和 2.7 $AP^m$。代码和预训练模型公开获取于:https://github.com/bok3948/GTSA

英文摘要:

Most invariance-based self-supervised methods rely on single object-centric images (e.g., ImageNet images) for pretraining, learning features that invariant to geometric transformation. However, when images are not object-centric, the semantics of the image can be significantly altered due to cropping. Furthermore, as the model becomes insensitive to geometric transformations, it may struggle to capture location information. For this reason, we propose a Geometric Transformation Sensitive Architecture designed to be sensitive to geometric transformations, specifically focusing on four-fold rotation, random crop, and multi-crop. Our method encourages the student to be sensitive by predicting rotation and using targets that vary with those transformations through pooling and rotating the teacher feature map. Additionally, we use patch correspondence loss to encourage correspondence between patches with similar features. This approach allows us to capture long-term dependencies in a more appropriate way than capturing long-term dependencies by encouraging local-to-global correspondence, which occurs when learning to be insensitive to multi-crop. Our approach demonstrates improved performance when using non-object-centric images as pretraining data compared to other methods that train the model to be insensitive to geometric transformation. We surpass DINO[Caron et al.[2021b]] baseline in tasks including image classification, semantic segmentation, detection, and instance segmentation with improvements of 4.9 $Top-1 Acc$, 3.3 $mIoU$, 3.4 $AP^b$, and 2.7 $AP^m$. Code and pretrained models are publicly available at: https://github.com/bok3948/GTSA

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。

↑