Semantic Image Segmentation

Data

• Original data : file01.png, file02.png, ...
• Labeled data : file01_L.png, file02_L.png

Class

• Total 11 class + Void class
• Pole, SignSymbol, Bicyclist, Pedestrian, Building, Fence, Pavement, Road, Car, Sky, Tree
• Class Distrubution

Model

• DeepLab v3 resnet 50

file

• run.py : Files that you run.
• model.py : Define Model Structure.
• labels.py : About pixel and label values in data.
• helpers.py : For calculation of iou, acc, loss, etc.

Test set result

• epoch 100, batch 2

  classes           IoU
  ---------------------
  Pole          : 0.671
  SignSymbol    : 0.872
  Bicyclist     : 0.896
  Pedestrian    : 0.829
  Building      : 0.959
  Fence         : 0.921
  Pavement      : 0.949
  Road          : 0.987
  Car           : 0.958
  Sky           : 0.941
  Tree          : 0.925
  ---------------------
  Mean IoU      : 0.901
  ---------------------
  loss : 0.1683 acc : 0.9686 miou : 0.9005

Environments

• CPU : Intel Core i9-10900X (10 Core / 20 Thread)
• RAM : Samsung 2666 (94G)
• GPU : GeForce RTX 2080 Ti (11G)

Member

• 김아경 : bzxz55@gmail.com
• 유소영 : yoosomail@gmail.com

reference

• hoya012