Skip to content

Latest commit

 

History

History
executable file
·
36 lines (24 loc) · 1.4 KB

Data_Augmentation.md

File metadata and controls

executable file
·
36 lines (24 loc) · 1.4 KB

GluonCv Semantic Segmentation Data Augmentation

Definitions :

  • Base_size: base_size is only used when resizing the image (for example, when augmenting, one possibility would be to resize the image anywhere between 0.5xbase_size and 2xbase_size)
  • Crop_size: called crop_size when augmenting the data, it's actually the final crop size of the image before it's passed to the network

Augmentation Process :

When augmenting your data using our custom data-augmentation function , multiple things might happen:

1- First possibility :

  • The image might be flipped left or right.
  • The image is resized anywhere between 0.5xbase_size and 2xbase_size.
  • If the width or height of the image are lower than the crop_size, black padding is added.
  • The image is then cropped anywhere to match the crop_size.
  • Blur might be added to the image.

2- Second Possibility:

  • The image might be resized to crop_size x crop_size and returned to the training network.

3- Third Possibility:

  • If the height or width of the image are lower than the crop_size,black padding is added.
  • The image is cropped to match crop_size, it's cropped either by using :
    • x=0 , y=0
    • x=width-crop_size, y=0
    • x=0, y=height-crop_size
    • x=width-crop_size, height-crop_size
  • Blur might be added to the image.

**4- ** Fourth Possibility:

  • If data_augmentation is false , the image is resized to crop_size x crop_size.