updated app

app.py

@@ -11,6 +11,7 @@ import tensorflow as tf
 from tensorflow import keras
 from tensorflow.keras import layers
 from keras.applications.vgg16 import preprocess_input
+from huggingface_hub import from_pretrained_keras
 
 # Image processing
 import PIL
@@ -23,19 +24,19 @@ filename_model_VGG16 = 'vgg16_best.keras'
 example_images_path = './example_images'
 
 # Loading model
-model = keras.models.load_model(path + filename_model_VGG16)
+model = from_pretrained_keras('ana-bernal/keras_15_dog_breed_eff')
 
 # Defining parameters
-breed_names_norm = [Chihuahua, papillon, beagle,
-                    Yorkshire_terrier, Australian_terrier,
-                    Scotch_terrier, golden_retriever,
-                    malinois, kelpie, Doberman, miniature_pinscher,
-                    Great_Dane, Pomeranian, standard_poodle, Mexican_hairless]
+breed_names_norm = ['Chihuahua', 'papillon', 'beagle',
+                    'Yorkshire_terrier', 'Australian_terrier',
+                    'Scotch_terrier', 'golden_retriever',
+                    'malinois', 'kelpie', 'Doberman', 'miniature_pinscher',
+                    'Great_Dane', 'Pomeranian', 'standard_poodle', 'Mexican_hairless']
 
 
 # Importing stopwords
-with open('./stopwords/stopwords.txt') as file:
-    my_stopwords = {line.rstrip() for line in file}
+# with open('./stopwords/stopwords.txt') as file:
+    # my_stopwords = {line.rstrip() for line in file}
 
 # Function definitions
 
@@ -51,7 +52,7 @@ def load_img_path(img_path, target_size, show=False):
 
     return img
 
-def predict(model, img, show=False):
+def classify_image(inp):
     """
     Returns a dictionnary: predicted_breeds, where the
     keys are [1,2,3] for the first, second and third more probable
@@ -61,36 +62,28 @@ def predict(model, img, show=False):
     Parameters:
         img: returned by the function load_img_path
     """
-    img_array = keras.preprocessing.image.img_to_array(img)
+    img_array = keras.preprocessing.image.img_to_array(inp)
     img_array = tf.expand_dims(img_array, 0)  # Creates a batch axis
 
-    predictions = model.predict(img_array, verbose=0)
-    scores = predictions[0]
+    predictions = model.predict(img_array, verbose=0).flatten()
+    confidences = {breed_names_norm[i]: float(predictions[i]) for i in range(15)}
 
-    # Keep the 3 more probable breeds
-    predicted_breeds = {}
-    for i in [1,2,3]:
-        idx = scores.argsort()[-i]  # First breed is the last proba when sorted
-        name = breed_names_norm[idx]
-        confidence = round(scores[idx]*100,2)
-        predicted_breeds[i] = {'idx':idx,'name':name,'confidence':confidence}
-       
-    return predicted_breeds
+    return confidences
 
 # --------------------------------------------------
 
 examples = [
-    ["Jquery/Javascript Opacity animation with scroll <p>I'm looking to change the opacity on an object (and have the transition be animated) based on a users scroll.\nexample(http://davegamache.com/)</p>\n\n<p>I've searched everywhere\nlike here, but it ends up pointing me to the waypoints plugin (http://stackoverflow.com/questions/6316757/opacity-based-on-scroll-position)</p>\n\n<p>I've implemented the [waypoints][1] plugin and have the object fading once it's higher than 100px. [Using the offet attribute] but would like to basically control the opacity of an object and have the animation be visible like the above example.</p>\n\n<p>I've searched all over- this is my last resort.\nAny help is greatly appreciated.</p>\n"],
-    ['Setting cross-domain cookies in Safari <p>I have to call domain A.com (which sets the cookies with http) from domain B.com.\nAll I do on domain B.com is (javascript): </p>\n\n<pre><code>var head = document.getElementsByTagName("head")[0];\nvar script = document.createElement("script");\nscript.src = "A.com/setCookie?cache=1231213123";\nhead.appendChild(script);\n</code></pre>\n\n<p>This sets the cookie on A.com on every browser I\'ve tested, except Safari.\nAmazingly this works in IE6, even without the P3P headers.</p>\n\n<p>Is there any way to make this work in Safari?</p>\n'],
-    ['Database migrations for SQL Server <p>I need a database migration framework for SQL Server, capable of managing both schema changes and data migrations.</p>\n\n<p>I guess I am looking for something similar to django\'s <a href="http://south.aeracode.org/" rel="noreferrer">South</a> framework here.</p>\n\n<p>Given the fact that South is tightly coupled with django\'s ORM, and the fact that there\'s so many ORMs for SQL Server I guess having just a generic migration framework, enabling you to write and execute in controlled and sequential manner SQL data/schema change scripts should be sufficient.</p>\n'],
+    ['example_images/01_test.jpg'],
+    ['example_images/02_test.jpg'],
+    ['example_images/03_test.jpg'],
+    ['example_images/04_test.jpg'],
 ]
 
-demo = gr.Interface(fn=tag_suggestion, 
-                    inputs="text", 
-                    outputs=["text"],
+demo = gr.Interface(fn=classify_image, 
+                    inputs=gr.Image(shape=(180, 180)), 
+                    outputs=gr.Label(num_top_classes=3),
                     examples=examples)
 
 
 if __name__ == "__main__":
-    demo.launch()
-
+    demo.launch()

requirements.txt

@@ -1,5 +1,4 @@
 gradio
 joblib
 tensorflow
-tensorflow_hub
-scikit-learn
+tensorflow_hub