arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2009.06469cs.CV
收藏

EfficientSeg:一个高效的语义分割网络

EfficientSeg: An Efficient Semantic Segmentation Network

Istanbul Technical University(伊斯坦布尔理工大学)

Vahit Bugra Yesilkaynak, Yusuf H. Sahin, Gozde Unal

摘要:

没有预先训练的权重和少量数据的深度神经网络训练需要更多的训练迭代。众所周知,对于语义分割任务,较深的模型比浅的模型更成功。因此,我们引入了EfficientSeg架构,这是U-Net的修改和可扩展版本,尽管它的深度很大,但它可以有效地训练。我们在Minicity数据集上评估了EfficientSeg架构,并使用相同的参数计数(51.5% mIoU)优于U-Net基线得分(40% mIoU)。我们最成功的模型获得了58.1%的mIoU分数,并在ECCV 2020 VIPriors挑战赛的语义分割赛道中获得第四名。

英文摘要:

Deep neural network training without pre-trained weights and few data is shown to need more training iterations. It is also known that, deeper models are more successful than their shallow counterparts for semantic segmentation task. Thus, we introduce EfficientSeg architecture, a modified and scalable version of U-Net, which can be efficiently trained despite its depth. We evaluated EfficientSeg architecture on Minicity dataset and outperformed U-Net baseline score (40% mIoU) using the same parameter count (51.5% mIoU). Our most successful model obtained 58.1% mIoU score and got the fourth place in semantic segmentation track of ECCV 2020 VIPriors challenge.

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。

↑