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 1806.08522cs.CV
收藏

基于渐进式搜索的高效语义分割

Efficient Semantic Segmentation using Gradual Grouping

IIIT-Hyderabad(海得拉巴国际信息技术学院);Texas Instruments(德州仪器)

Nikitha Vallurupalli, Sriharsha Annamaneni, Girish Varma, C V Jawahar, Manu Mathew, Soyeb Nagori

摘要:

用于语义分割的深度CNN具有高存储器和运行时间要求。已经提出了各种方法来使CNN高效,如分组,洗牌,深度可分离卷积。我们研究了这些技术在实时语义分割架构上的有效性,如ERFNet,以提高运行时间超过5倍。我们将这些技术部分或全部应用于CNN层,并在Cityscapes数据集上评估测试精度。我们获得了准确性与参数/FLOPs的权衡,为可以在指定的运行时预算下运行的模型提供了准确性分数。我们进一步提出了一种新的训练过程,它从密集卷积开始,但逐渐向分组卷积发展。我们表明,我们提出的训练方法和高效的架构设计可以将精度提高8%以上,将深度可分离卷积应用于ERFNet的编码器并附加一个轻量级解码器。这导致模型的FLOP提高了5倍,而相对于ERFNet,准确性仅下降了4%。

英文摘要:

Deep CNNs for semantic segmentation have high memory and run time requirements. Various approaches have been proposed to make CNNs efficient like grouped, shuffled, depth-wise separable convolutions. We study the effectiveness of these techniques on a real-time semantic segmentation architecture like ERFNet for improving run time by over 5X. We apply these techniques to CNN layers partially or fully and evaluate the testing accuracies on Cityscapes dataset. We obtain accuracy vs parameters/FLOPs trade offs, giving accuracy scores for models that can run under specified runtime budgets. We further propose a novel training procedure which starts out with a dense convolution but gradually evolves towards a grouped convolution. We show that our proposed training method and efficient architecture design can improve accuracies by over 8% with depth wise separable convolutions applied on the encoder of ERFNet and attaching a light weight decoder. This results in a model which has a 5X improvement in FLOPs while only suffering a 4% degradation in accuracy with respect to ERFNet.

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。

↑