CLASSIFICATION DATA AUGMENTATION
Data augmentation is a crucial strategy when dealing with image classification. GILA uses the most common data augmentation techniques that Keras provides:
Width shift
Shift range in the horizontal axis as a percentage of the image width:
-da_width 10
default value=0
Height shift
Shift range in the vertical axis as a percentage of the image height:
-da_height 20
default value=0
Rotation
Maximum rotation angle
-da_rotaion 10
default value=0
Zoom
Zoom multiplier range applied to the image [1-zoom,1+zoom]
-da_zoom 0.1
default value=0.0
Horizontal Flip
Flipt the image horizontally
-da_flip_h
default value=no
Vertical Flip
Flip the image vertically
-da_flip_v
default value=no
Not only for images:
Gaussian noise
Additive gaussian noise to all the activation layers
-da_gauss 0.3
default value=0.0
Rescale
A fixed scale of the input values (1/value). Typically value=255 for images
-da_rescale 255
All these data augmentation can be combined:
python3 gila.py -trdir train -tsdir test -width 32 -height 32 -mode class -model resnet50 -da_width 20 -da_height 20 -da_flip_h -da_zoom 0.2 -da_rescale 255.0