CLASSIFICATION DATA SOURCES

Images can be specified in two different ways:

List Files:

A file with the name and class of each image. For example:

tr.txt:

train/airplane/10008_airplane.png 0
train/automobile/10839_automobile.png 1
train/bird/12362_bird.png 2
train/cat/1316_cat.png 3
train/deer/1297_deer.png 4
train/dog/13121_dog.png 5
train/frog/12885_frog.png 6
train/horse/12633_horse.png 7
train/ship/19097_ship.png 8
train/truck/12979_truck.png 9

Files can be everywhere, it is not necessary that images of the same class to be in the same folder. Classes start in 0.

These files will be defined using the “-trfile” and “-tsfile” arguments:

python3 gila.py -mode class -trfile tr.txt -tsfile ts.txt

---

Directories:

This method is more convenient because images are not loaded in memory but processed in a flow manner.

Each class is a directory containing all the images of that class. In this case, images of the same class must be in the same folder. And the folders of the different classes must be in the same directory for training or test:

train/
  -- airpale/
  -- automobile/
  -- bird/
  -- cat/
  -- deer/
  -- dog/
  -- frog/
  -- horse/
  -- ship/
  -- truck/

test/
  -- airpale/
  -- automobile/
  -- bird/
  -- cat/
  -- deer/
  -- dog/
  -- frog/
  -- horse/
  -- ship/
  -- truck/

These folders will be defined using the “-trdir” and “-tsdir” arguments:

python3 gila.py -mode class -trdir train -tsdir test

---

Image size

No matter the size of the images GILA will resize them to the working size specified with the -width and -height arguments:

-width 64 -height 64

default=240x240

To rescale the images GILA can use a resize or crop:

-resize {resize,crop}

default=resize

Also we can define the number of channels:

-chan {rgb,gray}  

default=rgb

Next