Merge branch 'master' into bugfix/dynamic-num-letters

# Conflicts:
#	src/create_train_data.py

Readme.md

@@ -23,16 +23,12 @@ $ npm i
 ```
 |_data
       |_captcha
-          |_ captcha_1.jpg
-          |_ captcha_2.jpg
-```
-* Create mapping file `data/captcha.json` to map your train image with corresponding label
-```json
-{
-    "captcha_1.jpg": "HEYMEN",
-    "captcha_2.jpg": "XINCHA"
-}
+          |_ xss7.jpg
+          |_ tvu4.jpg
 ```
+**IMPORTANT:** Note each image file is named with it's own solution.
+
+That means that if an image is named `A1bD3.jpg`, it corresponds to a captcha's whose solution is `A1bD3`  
 
 #### Build train data for model
 Run `src/create_train_data.py` will save your train data as `data/captcha.npz` compressed file.

requirements.txt

@@ -20,7 +20,7 @@ mock==2.0.0
 networkx==2.1
 numpy==1.14.5
 pbr==4.0.4
-Pillow==5.1.0
+Pillow==6.2.0
 protobuf==3.6.0
 pyparsing==2.2.0
 python-dateutil==2.7.3
@@ -31,7 +31,7 @@ scipy==1.1.0
 six==1.11.0
 subprocess32==3.5.2
 tensorboard==1.8.0
-tensorflow==1.8.0
+tensorflow==1.12.2
 termcolor==1.1.0
 toolz==0.9.0
-Werkzeug==0.14.1
+Werkzeug==0.15.3

src/create_train_data.py

@@ -2,11 +2,22 @@
 import os
 import json
 from skimage import io
+from skimage.color import rgb2gray
 from img import split_letters
 import numpy as np
 
+
+# Helper methods
+def strip_extension(filename):
+    return filename[:filename.rindex('.')]
+
+
+def build_data_map(data_path):
+    files = os.listdir(data_path)
+    return {x: strip_extension(x) for x in files}
+
+
 DATA_DIR = 'data'
-DATA_MAP = os.path.join(DATA_DIR, 'captcha.json')
 DATA_FULL_DIR = os.path.join(DATA_DIR, 'captcha')
 DATA_TRAIN_DIR = os.path.join(DATA_DIR, 'train')
 DATA_TRAIN_FILE = os.path.join(DATA_DIR, 'captcha')
@@ -15,21 +26,21 @@
 data_x = []
 data_y = []
 
-# load image content json file
-with open(DATA_MAP) as f:
-    image_contents = json.load(f)
+# build image contents map
+image_contents = build_data_map(DATA_FULL_DIR)
 
 # load image and save letters
 counter = 0
+
 for fname, contents in image_contents.iteritems():
     counter += 1
     print(counter, fname, contents)
-    image = io.imread(os.path.join(DATA_FULL_DIR, fname))
+    original_image = io.imread(os.path.join(DATA_FULL_DIR, fname))
+    grayscale_image = rgb2gray(original_image)
 
     # split image
-    letters = split_letters(image, num_letters=len(contents), debug=True)
+    letters = split_letters(grayscale_image, num_letters=len(contents), debug=True)
     if letters != None:
-        fname = fname.replace('.jpg', '.png')
         for i, letter in enumerate(letters):
             content = contents[i]
             # add to dataset
@@ -40,7 +51,7 @@
             fpath = os.path.join(DATA_TRAIN_DIR, content)
             if not os.path.exists(fpath):
                 os.makedirs(fpath)
-            letter_fname = os.path.join(fpath, str(i+1) + '-' + fname)
+            letter_fname = os.path.join(fpath, str(i+1) + '-' + strip_extension(fname) + '.png')
             io.imsave(letter_fname, 255 - letter) # invert black <> white color
     else:
         print('Letters is not valid')