eduzhai > Applied Sciences > Engineering >

Few-Shot Semantic Segmentation Augmented with Image-Level Weak Annotations

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 10 pages

Abstract: Despite the great progress made by deep neural networks in the semanticsegmentation task, traditional neural network-based methods typically sufferfrom a shortage of large amounts of pixel-level annotations. Recent progress infew-shot semantic segmentation tackles the issue by utilizing only a fewpixel-level annotated examples. However, these few-shot approaches cannoteasily be applied to utilize image-level weak annotations, which can easily beobtained and considerably improve performance in the semantic segmentationtask. In this paper, we advance the few-shot segmentation paradigm towards ascenario where image-level annotations are available to help the trainingprocess of a few pixel-level annotations. Specifically, we propose a newframework to learn the class prototype representation in the metric space byintegrating image-level annotations. Furthermore, a soft masked average poolingstrategy is designed to handle distractions in image-level annotations.Extensive empirical results on PASCAL-5i show that our method can achieve 5.1 and 8.2 increases of mIoU score for one-shot settings with pixel-level andscribble annotations, respectively.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×