transfer learning example

Author: lukas

examples/keras-transfer-learning/elephant.jpg

@@ -0,0 +1,47 @@
+from subprocess import call
+import os
+from urllib.request import urlretrieve
+from keras.preprocessing.image import load_img, img_to_array
+import numpy as np
+
+dataset_url = "http://aisdatasets.informatik.uni-freiburg.de/" \
+              "freiburg_groceries_dataset/freiburg_groceries_dataset.tar.gz"
+
+def download():
+    print("Downloading dataset.")
+    urlretrieve(dataset_url, "freiburg_groceries_dataset.tar.gz")
+    print("Extracting dataset.")
+    call(["tar", "-xf", "freiburg_groceries_dataset.tar.gz", "-C", "."])
+    os.remove("freiburg_groceries_dataset.tar.gz")
+    print("Done.")
+
+    
+def load_data():
+    if (not os.path.exists("images")):
+        download()
+        
+    x_train = []
+    y_train = []
+    x_test = []
+    y_test = []
+    class_names = []
+    category_num = 0
+    
+    for category in sorted(os.listdir("images")):
+        class_names.append(category)
+        count = 0
+        for img in sorted(os.listdir(os.path.join("images", category))):
+            if (not img.endswith(".png")):
+                continue
+                
+            x = load_img(os.path.join("images", category, img),target_size=(224, 224))
+            if count < 10:
+                x_test.append(img_to_array(x))
+                y_test.append(category_num)
+            else:
+                x_train.append(img_to_array(x))
+                y_train.append(category_num)
+            count += 1
+        category_num += 1
+    return (np.array(x_train), np.array(y_train)), (np.array(x_test), np.array(y_test)), class_names
+    

examples/keras-transfer-learning/groceries.py

@@ -0,0 +1,5 @@
+[default]
+entity = bloomberg-class
+project = groceries
+base_url = https://api.wandb.ai
+