eduzhai > Applied Sciences > Engineering >

AinnoSeg Panoramic Segmentation with High Perfomance

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 6 pages

Abstract: Panoramic segmentation is a scene where image segmentation tasks is moredifficult. With the development of CNN networks, panoramic segmentation taskshave been sufficiently developed.However, the current panoramic segmentationalgorithms are more concerned with context semantics, but the details of imageare not processed enough. Moreover, they cannot solve the problems whichcontains the accuracy of occluded object segmentation,little objectsegmentation,boundary pixel in object segmentation etc. Aiming to address theseissues, this paper presents some useful tricks. (a) By changing the basicsegmentation model, the model can take into account the large objects and theboundary pixel classification of image details. (b) Modify the loss function sothat it can take into account the boundary pixels of multiple objects in theimage. (c) Use a semi-supervised approach to regain control of the trainingprocess. (d) Using multi-scale training and reasoning. All these operationsnamed AinnoSeg, AinnoSeg can achieve state-of-art performance on the well-knowndataset ADE20K.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×