one big commit, sorry

common/backend-utils/xaidemo/tracking/config.py

@@ -5,7 +5,7 @@
 
 class TrackingSettings(BaseSettings):
     experiment: bool = False
-    service_name: str
+    service_name: str = "tracking_service"
     collector_url: Optional[str]
     collector_timeout: int = 60
 

guess-the-country/country-backend/country/config.py

@@ -2,7 +2,7 @@
 
 
 class Settings(BaseSettings):
-    service_name: str = "country-service"
+    service_name = "country-service"
     root_path: str = ""
     path_prefix: str = ""
     # Google Maps API access

guess-the-country/country-backend/country/explainer/explain.py

@@ -19,9 +19,9 @@ class Explanation(BaseModel):
 def explain(data):
     encoded_data = str(data)
     image = load_image(encoded_data)
-    pre_image = preprocess(img=image)
+    preproc_image = preprocess(img=image)
 
-    explanation = explain_cnn(pre_image, model)
+    explanation = explain_cnn(np.array(image), preproc_image, model)
     explanation_id = uuid.uuid4()
 
     encoded_image_string = convert_explanation(explanation)
@@ -45,12 +45,18 @@ def convert_explanation(explanation):
 
 
 @traced
-def explain_cnn(image, model_=model):
-    segment_mask, segment_weights = explain_classification(image=image,
+def explain_cnn(image, preproc_image, model_=model):
+    try:
+        segment_mask, segment_weights = explain_classification(image=preproc_image,
                                                            segmentation_method="felzenszwalb",
                                                            segmentation_settings={},
-                                                           predict_fn=model_.predict_,
-                                                           num_of_samples=500,
-                                                           p=0.9)
+                                                           predict_fn=model_.predict,
+                                                           num_of_samples=100,
+                                                           p=0.5)
+
+        return render_explanation(preproc_image, segment_mask, segment_weights, 
+                                  positive="violet", coverage=0.15, opacity=0.5)
+
+    except ZeroDivisionError:
+        return image
 
-    return render_explanation(image, segment_mask, segment_weights, positive="violet", coverage=0.15, opacity=0.5)

guess-the-country/country-backend/country/main.py

@@ -2,7 +2,7 @@
 import sys
 
 from fastapi import FastAPI
-from xaidemo import tracing, tracking
+from xaidemo import tracing#, tracking
 
 
 from xaidemo.routers import vue_frontend
@@ -16,7 +16,7 @@
 tracing.set_up()
 
 app = FastAPI(root_path=settings.root_path)
-tracking.instrument_app(app)
+#tracking.instrument_app(app)
 app.include_router(api, prefix=settings.path_prefix)
 app.include_router(vue_frontend(__file__), prefix=settings.path_prefix)
 

guess-the-country/country-backend/country/model/predict.py

@@ -35,7 +35,7 @@ def load_image(encoded_data):
 
 @traced
 def predict_image(image):
-    prediction = model.predict(image)
+    prediction = process_predict(image[None,:,:,:])
     result = decode_model_output(prediction)
     return result
 
@@ -44,9 +44,21 @@ def preprocess(img):
     img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
     # resize image to match model's expected sizing
     img_resize = cv2.resize(img_rgb, (IMG_SIZE, IMG_SIZE))
-    image = tf.keras.applications.mobilenet_v2.preprocess_input(img_resize)
-    pre_image = image.reshape(-1, IMG_SIZE, IMG_SIZE, 3)
-    return pre_image
+    #image = tf.keras.applications.mobilenet_v2.preprocess_input(img_resize)
+    #pre_image = img_resize
+    #pre_image = image.reshape(-1, IMG_SIZE, IMG_SIZE, 3)
+
+    return img_resize
+
+
+def process_predict(batch: np.ndarray): 
+    """
+    input: np array of size (batch, height, width, 3)
+    output: np array of size (batch, 4) where every row represents the softmax output of a
+            image inside the given batch
+    """
+    proc_batch = tf.keras.applications.mobilenet_v2.preprocess_input(batch)
+    return model.predict(proc_batch)
 
 
 MODEL_OUTPUT_MAP = ["Tel_Aviv", "Westjerusalem", "Berlin", "Hamburg"]

guess-the-country/country-backend/country/streetview/collect.py

@@ -1,8 +1,14 @@
 import base64
+import io
+import json
 import random
+import os
+
 from urllib.error import URLError
 from urllib.request import urlopen
 
+from itertools import cycle
+from PIL import Image
 from pydantic import BaseModel
 from shapely.geometry import Polygon
 from xaidemo.http_client import AioHttpClientSession
@@ -194,40 +200,28 @@ class Streetview(BaseModel):
 
 country_array = [tel_aviv, jerusalem, berlin, hamburg]
 
+with open('/country/streetview/filename2class.json', 'r') as f:
+    filename2class = json.load(f)
+
+
+filenames = [f for f in filename2class] 
+random.shuffle(filenames)
+file_iter = cycle(filenames)
+
 
 @traced
 async def get_streetview(API_KEY):
-    async with AioHttpClientSession() as session:
-        nominated_country = random.randint(0, 3)
-        poly = country_array[nominated_country]['polygon']
-        status = False
-        while status != 'OK':
-            coord = generate_random(poly)
-            lng = coord[0][0]
-            lat = coord[0][1]
-            locstring = str(lat) + "," + str(lng)
-            try:
-                async with session.get(
-                        API_URL + "?key=" + API_KEY + "&location=" + locstring + "&source=outdoor") as response:
-                    json_body = (await response.json())
-                    status = json_body['status']
-                    print(status)
-                    if status == 'REQUEST_DENIED':
-                        print("NO API-KEY is definied, please set environment variable GOOGLE_MAPS_API_TOKEN")
-                        break
-            except AioHttpClientSession.exceptions.TimeoutError:
-                print(AioHttpClientSession.exceptions.TimeoutError)
-        print("    ========== Got one! ==========")
-        url = GOOGLE_URL + API_KEY + "&location=" + locstring
-        try:
-            contents = urlopen(url).read()
-            # urlretrieve(url, outfile)
-        except URLError:
-            print(URLError)
-        status = False
-        encoded_image_string = base64.b64encode(contents)
-        encoded_bytes = bytes("data:image/png;base64,",
+    filename = next(file_iter)
+    try: 
+        image = Image.open(f"/country/streetview/streetview_images/{filename}")
+    except:
+        raise Exception(f'FILE NOT FOUND!')
+
+    buffered = io.BytesIO()
+    image.save(buffered, format="JPEG")
+    encoded_image_string = base64.b64encode(buffered.getvalue())
+    encoded_bytes = bytes("data:image/png;base64,",
                               encoding="utf-8") + encoded_image_string
     return Streetview(
         image=encoded_bytes,
-        class_label=country_array[nominated_country]['city'])
+        class_label=filename2class[filename])