1
InstanceCut: from Edges to Instances with MultiCut A. Kirillov, E. Levinkov, B. Andres, B. Savchynskyy, and C. Rother (ARXIV 2016) I Instance-aware semantic segmentation I Challenges: I Meaningless labels (car number 5) I Varying number of objects I Set of pixels vs. bounding boxes I Need for a more refined labelling of the training data (rare classes) I An instance-agnostic semantic segmentation using CNNs I All instance-boundaries using a new instance-aware edge detection model I Combined into a novel MultiCut formulation I Evaluated on CityScapes: particularly well for rare object classes I Not handled: instances that are formed by disconnected regions in the image

InstanceCut: from Edges to Instances with MultiCutInstanceCut: from Edges to Instances with MultiCut A. Kirillov, E. Levinkov, B. Andres, B. Savchynskyy, and C. Rother (ARXIV 2016)

  • Upload
    others

  • View
    4

  • Download
    0

Embed Size (px)

Citation preview

Page 1: InstanceCut: from Edges to Instances with MultiCutInstanceCut: from Edges to Instances with MultiCut A. Kirillov, E. Levinkov, B. Andres, B. Savchynskyy, and C. Rother (ARXIV 2016)

InstanceCut: from Edges to Instances with MultiCutA. Kirillov, E. Levinkov, B. Andres, B. Savchynskyy, and C. Rother (ARXIV 2016)

I Instance-aware semantic segmentation

I Challenges:I Meaningless labels (car number 5)I Varying number of objectsI Set of pixels vs. bounding boxesI Need for a more refined labelling of the training data (rare classes)

I An instance-agnostic semantic segmentation using CNNs

I All instance-boundaries using a new instance-aware edge detection model

I Combined into a novel MultiCut formulation

I Evaluated on CityScapes: particularly well for rare object classes

I Not handled: instances that are formed by disconnected regions in the image