Graph Laplacian Regularization based on the Differences of Neighboring Pixels for Conditional Convolutions for Instance Segmentation
Graph Laplacian Regularization based on the Differences of Neighboring Pixels for Conditional Convolutions for Instance Segmentation
复制标题
DOI：
10.1109/icpr56361.2022.9956326
复制
发表时间：
2022-08
期刊：
影响因子：
--
通讯作者：
Shinji Uchinoura;Takio Kurita
中科院分区：
文献类型：
--
作者：
Shinji Uchinoura;Takio Kurita
我们提出了一种简单有效的实例分割正则化方法GLRDN-L2（基于相邻像素差异的图拉普拉斯正则化）。实例分割是计算机视觉中的一个具有挑战性的任务。多年来，基于roi的方法（如Mask R-CNN）一直是表现最好的方法；然而，最近提出的CondInst，它采用动态fcn作为掩码头并执行实例感知掩码预测，优于mask R-CNN。据我们所知，所有的方法都是基于像素损失（如Dice Loss）来优化模型的。即使使用高分辨率的掩模，在实例中也存在边界模糊和空洞等问题。我们假设这些问题是由于邻近像素之间的关系中包含的空间结构和上下文信息没有很好地纳入模型。为了解决这些问题，我们提出了一种正则化方法，使用由相邻像素之间的差异组成的图来惩罚空间结构中的错误。我们比较了在CondInst中使用正则化和不使用正则化训练的模型，以验证自然扩展的效果，该扩展为损失函数添加了微分，并展示了在COCO和cityscape数据集上的性能改进。
We propose a simple and effective regularization method for instance segmentation, GLRDN-L2 (Graph Laplacian Regularization based on Differences of Neighboring Pixels). Instance segmentation is a challenging task in computer vision. For many years, ROI-based methods such as Mask R-CNN have dominantly presented the top performances; however, the recently proposed CondInst, which employs dynamic FCNs as a mask head and performs instance-aware mask prediction, outperforms Mask R-CNN. To our best knowledge, all methods optimize a model based on pixel-wise losses such as Dice Loss. Even with the results of high-resolution masks, there are problems such as blurred boundaries and hollows in the instances. We assume that these problems are due to the spatial structure and contextual information contained in the relationships between neighboring pixels not being incorporated well in the model. To address these problems, we propose a regularization that penalizes the errors in the spatial structure with a graph composed of the differences between neighboring pixels. We compare models trained with and without our regularization in CondInst to validate the effect of a natural extension that adds differentiation to the loss function and demonstrate performance improvement on both COCO and Cityscapes datasets.