Rearrange images into subfolders

First thing we need to do is rearrange the "raw" dataset by saving all the images belonging to each class in their own subfolder.

This rearrangement is needed in order to allow us make use of the amazing Kera's ImageDataGenerator



In [1]:

    
import os



In [2]:

    
%mkdir "train/dog"
%mkdir "train/cat"



In [3]:

    
for dir, subdir, files in os.walk("train"):
    if len(subdir) == 0:
        continue
    for file in files:
        category = file.split(".")[0]
        os.rename("{}/{}".format(dir,file), "{}/{}/{}".format(dir, category, file))

Make validation set



In [17]:

    
%mkdir "valid/dog"
%mkdir "valid/cat"



In [8]:

    
dogs = [x for x in os.listdir("train/dog")]



In [9]:

    
cats = [x for x in os.listdir("train/cat")]



In [14]:

    
print len(cats)
len(dogs)



In [15]:

    
import random



In [18]:

    
if len(os.listdir("valid/dog")) < 1:
    for n in random.sample(range(len(dogs)), 1000):
        os.rename("train/dog/{}".format(dogs[n]), "valid/dog/{}".format(dogs[n]))



In [19]:

    
if len(os.listdir("valid/cat")) < 1:
    for n in random.sample(range(len(cats)), 1000):
        os.rename("train/cat/{}".format(cats[n]), "valid/cat/{}".format(cats[n]))



In [ ]: