Merge branch 'master' of github.com:nardeas/glados

cnn_model/app.py

@@ -0,0 +1,33 @@
+from flask import Flask, request, jsonify
+from utils import crop_frames, get_crop_sign, store_img, process_image
+from cnn_model import predict
+import numpy as np
+import time
+app = Flask(__name__)
+
+@app.route('/predict/<id>', methods=['GET'])
+def predict_sign(id):
+    full_path = "/tmp/{0}-full.jpg".format(id)
+    crop_path = "/tmp/{0}".format(id)
+    cropped_img = get_crop_sign(full_path)
+
+    if not cropped_img:
+        return jsonify({
+            'image': None,
+            'valid': False
+        })
+
+    store_img(path=crop_path, image=cropped_img)
+    label = predict(np.array(process_image(cropped_img)))
+
+    data = {'id': id,
+            'timestamp': int(time.time()),
+            'image': crop_path,
+            'image_full': full_path,
+            'probability': 0.0,
+            'valid': label[0]}
+
+    return jsonify(data)
+
+if __name__ == '__main__':
+    app.run(port=8080)

cnn_model/cnn_model.py

@@ -4,6 +4,7 @@
 
 from utils import read_data
 from sklearn.model_selection import train_test_split
+import numpy as np
 
 data = read_data()
 X_train = data["x"]
@@ -22,7 +23,7 @@
 def conv_net(features, reuse, n_classes, is_training):
     # Define a scope for reusing the variables
     with tf.variable_scope('ConvNet', reuse=reuse):
-        x = tf.placeholder(tf.float32, shape=(None, 32, 32, 3))
+        x = tf.placeholder(tf.float32, shape=(None, 32, 32, 3), name="feature_input")
         x = features['images']
 
         # Layer 1: Convolutional. Input = 32x32x3. Output = 28x28x32.
@@ -35,15 +36,11 @@ def conv_net(features, reuse, n_classes, is_training):
 
         # Fully connected layer (in tf contrib folder for now)
         out = tf.layers.dense(fc1, units=n_classes, name='out')
-    return out, x
+    return out
 
-logits_train_x = None
-logits_test_x = None
 def model_fn(features, labels, mode):
-    global logits_train_x
-    global logits_test_x
-    logits_train, logits_train_x = conv_net(features, False, n_classes=num_classes, is_training=True)
-    logits_test, logits_test_x = conv_net(features, True, n_classes=num_classes, is_training=False)
+    logits_train = conv_net(features, False, n_classes=num_classes, is_training=True)
+    logits_test = conv_net(features, True, n_classes=num_classes, is_training=False)
 
     # Predictions
     pred_classes = tf.argmax(logits_test, axis=1)
@@ -63,6 +60,7 @@ def model_fn(features, labels, mode):
     # Evaluate the accuracy of the model
     acc_op = tf.metrics.accuracy(labels=labels, predictions=pred_classes)
 
+
     # TF Estimators requires to return a EstimatorSpec, that specify
     # the different ops for training, evaluating, ...
     estim_specs = tf.estimator.EstimatorSpec(
@@ -81,43 +79,18 @@ def model_fn(features, labels, mode):
 input_fn = tf.estimator.inputs.numpy_input_fn(
     x={'images': X_train}, y=y_train,
     batch_size=batch_size, num_epochs=None, shuffle=True)
+
 # Train the Model
-model.train(input_fn, steps=num_steps)
+#model.train(input_fn, steps=num_steps)
 
-# Evaluate the Model
-# Define the input function for evaluating
-input_fn = tf.estimator.inputs.numpy_input_fn(
-    x={'images': X_train}, y=y_train,
-    batch_size=batch_size, shuffle=False)
-# Use the Estimator 'evaluate' method
-e = model.evaluate(input_fn)
+def predict(image_array):
+    input_fn = tf.estimator.inputs.numpy_input_fn(x={'images': np.array([image_array], dtype=np.float32)}, num_epochs=1, shuffle=False)
+    return list(model.predict(input_fn))
 
 validation_data = read_data(path='validation/', true_label=None)
 X_val = validation_data['x']
 y_val = validation_data['y']
 
-# Evaluate the Model
-# Define the input function for evaluating
-input_fn = tf.estimator.inputs.numpy_input_fn(
-    x={'images': X_val}, y=y_val,
-    batch_size=batch_size, shuffle=False)
-# Use the Estimator 'evaluate' method
-e = model.evaluate(input_fn)
-print("Stop signs", e)
-# Validate model
-# input_fn = tf.estimator.inputs.numpy_input_fn(x={'images': X_val}, num_epochs=1, shuffle=False)
-# v = list(model.predict(input_fn))
-# print("Predicted stop signs", v)
-# print("Testing Accuracy:", e['accuracy'])
-
-validation_data = read_data(path='validation_false/', true_label="asd")
-X_val = validation_data['x']
-y_val = validation_data['y']
-
-input_fn = tf.estimator.inputs.numpy_input_fn(
-    x={'images': X_val}, y=y_val,
-    batch_size=batch_size, shuffle=False)
-# Use the Estimator 'evaluate' method
-e = model.evaluate(input_fn)
-print("No stop signs", e)
-
+validation_data_false = read_data(path='validation_false/', true_label="asd")
+X_val_false = validation_data_false['x']
+y_val_false = validation_data_false['y']

cnn_model/requirements.txt

@@ -1,18 +1,22 @@
-appnope==0.1.0
 backports.functools-lru-cache==1.4
 backports.shutil-get-terminal-size==1.0.0
 backports.weakref==1.0.post1
 bleach==1.5.0
+click==6.7
 cycler==0.10.0
 decorator==4.1.2
 enum34==1.1.6
+Flask==0.12.2
 funcsigs==1.0.2
 futures==3.1.1
 html5lib==0.9999999
 ipdb==0.10.3
 ipython==5.5.0
 ipython-genutils==0.2.0
+itsdangerous==0.24
+Jinja2==2.10
 Markdown==2.6.9
+MarkupSafe==1.0
 matplotlib==2.1.0
 mock==2.0.0
 mutable==0.2.2

cnn_model/trained_model/graph.pbtxt

@@ -982,7 +982,7 @@ node {
   }
 }
 node {
-  name: "ConvNet/Placeholder"
+  name: "ConvNet/feature_input"
   op: "Placeholder"
   input: "^add"
   attr {
@@ -3580,7 +3580,7 @@ node {
   }
 }
 node {
-  name: "ConvNet_1/Placeholder"
+  name: "ConvNet_1/feature_input"
   op: "Placeholder"
   input: "^add"
   attr {
@@ -14003,7 +14003,7 @@ node {
         dtype: DT_STRING
         tensor_shape {
         }
-        string_val: "_temp_ab0101c4d73346ed92c4b54680eaf431/part"
+        string_val: "_temp_0856d67a756543af91b0ed2baec2438f/part"
       }
     }
   }

cnn_model/utils.py

@@ -17,7 +17,7 @@ def read_data(path='GTSRB/Final_Training/Images/*/', shape=(32, 32), true_label=
             if true_label is None:
                 print(filename)
 
-            if label is not 1 and i > 150:
+            if label is not 1 and i > 200:
                 break
 
             im = Image.open(filename)
@@ -31,8 +31,18 @@ def read_data(path='GTSRB/Final_Training/Images/*/', shape=(32, 32), true_label=
 
     return {"x": np.array(X, dtype=np.float32), "y": np.array(Y, dtype=np.float32)}
 
+def process_image(im, shape=(32, 32)):
+    im = im.convert('RGB')
+    return im.resize(shape, Image.LANCZOS)
 
-def crop_sign(img_path):
+def get_image_array(path):
+    im = Image.open(path)
+    return np.array(process_image(im))
+
+def store_img(path, image, format="jpg"):
+    image.save(path + '.' + format)
+
+def get_crop_sign(img_path):
     img = cv2.imread(img_path)
     original_img = img.copy()
     final_img = Image.open(img_path)
@@ -120,7 +130,7 @@ def crop_sign(img_path):
 
 def crop_frames(filenames):
     for i, file in enumerate(filenames):
-        image = crop_sign(file)
+        image = get_crop_sign(file)
         if image:
             image.save("crop" + str(i) + '.bmp')
         else:

docker/server/Dockerfile

@@ -2,7 +2,6 @@ FROM glados:tf
 
 # Setup env
 ENV NODE_VERSION=8.6.0 NPM_VERSION=5
-ENV RM_DIRS=/usr/include
 
 # Install node & npm
 RUN echo "Collecting dependencies..." && \
@@ -49,7 +48,7 @@ RUN echo "Collecting dependencies..." && \
   \
   echo "Cleaning up..." && \
   \
-  rm -rf ${RM_DIRS} /node-v${NODE_VERSION}* /usr/share/man /tmp/* \
+  rm -rf /node-v${NODE_VERSION}* /usr/share/man /tmp/* \
     /root/.npm /root/.node-gyp /root/.gnupg /usr/lib/node_modules/npm/man \
     /usr/lib/node_modules/npm/doc /usr/lib/node_modules/npm/html /usr/lib/node_modules/npm/scripts && \
   echo "Version info - node=$(node -v), npm=$(npm -v)"
@@ -61,9 +60,15 @@ RUN pip install opencv-python
 WORKDIR /opt
 COPY ./entry.sh /opt/entry.sh
 COPY ./pkg.tar.gz /opt/pkg.tar.gz
+COPY ./cnn.tar.gz /opt/cnn.tar.gz
 
 # Unpack
-RUN tar -zxvf /opt/pkg.tar.gz && \
+RUN tar -zxvf /opt/cnn.tar.gz && \
+    tar -zxvf /opt/pkg.tar.gz
+
+RUN cd /opt/cnn_model && \
+    pip install -r requirements.txt && \
+    \
     cd /opt/server && \
     npm i && \
     chmod a+x /opt/* && \

server/geo.js

@@ -0,0 +1,30 @@
+// Returns radians from degrees
+const radians = d => (d * Math.PI / 180);
+
+// Returns degrees from radians
+const degrees = r => (r * 180 / Math.PI);
+
+// Returns the haversine distance in km between two coords
+const distance = (lat1, lng1, lat2, lng2) => {
+
+  // Get deltas
+  const dLat = radians(lat2-lat1);
+  const dLon = radians(lng2-lng1);
+
+  // Convert to radians
+  lat1 = radians(lat1);
+  lat2 = radians(lat2);
+
+  const a = (Math.sin(dLat/2) * Math.sin(dLat/2) +
+            (Math.sin(dLon/2) * Math.sin(dLon/2) *
+             Math.cos(lat1) * Math.cos(lat2)));
+
+  return 6371 * 2 * Math.atan2(Math.sqrt(a), Math.sqrt(1-a));
+}
+
+// Export all
+module.exports = {
+  radians,
+  degress,
+  distance,
+}

server/index.js

@@ -1,10 +1,18 @@
+const fs = require('fs');
+const axios = require('axios');
 const server = require('./server');
 const http = require('./http');
 
+// Base path
+const { BASE_PATH = 'localhost:8000' } = process.env;
+
 // Collect api and streams
 const { sendEvent } = server;
 const { connections, messages, errors, options } = server;
 
+// Create history storage
+const history = {};
+
 // New connections
 connections.subscribe(([ws, chan]) => {
   console.log(`* [chan:${chan}] new connection ${ws._socket.remoteAddress}`);
@@ -17,9 +25,62 @@ messages.subscribe(([evt, chan]) => {
 
 // Event: capture image
 messages
-.filter(([evt, _]) => evt.type === 'MARK_CAPTURE')
+.filter(([evt, _]) => evt.type === 'MARK_SEND')
 .subscribe(([evt, chan]) => {
-  console.log(evt.payload);
+
+  const { payload } = evt;
+
+  // IN:
+  //   {
+  //     id: 4575917962,
+  //     timestamp: 1511649981213,
+  //     lat: 60.299,
+  //     lng: 24.2119,
+  //     data: <base64>
+  //   }
+
+  // Let's hack it together
+  if (payload.data) {
+
+    // Create filename
+    const filename = `/tmp/${payload.id}-full.jpg`;
+
+    // Write base64 data to JPG file
+    fs.writeFile(filename, payload.data, 'base64', err => {
+      if (err) {
+        return console.log(err);
+      }
+
+      // Log new image
+      console.log(`* got new image ${filename}`);
+
+      // Try predict
+      axios
+      .get(`http://localhost:8080/predict/${payload.id}`)
+      .then(res => {
+
+        console.log(res.data);
+
+        if (res.data.valid) {
+          // Respond
+          sendEvent(chan, 'MARK_RECEIVE', {
+            id: payload.id,
+            timestamp: +new Date(),
+            lat: payload.lat,
+            lng: payload.lng,
+            image: `http://${BASE_PATH}/static/${payload.id}-full.jpg`,
+            image_full: `http://${BASE_PATH}/static/${payload.id}-full.jpg`,
+            probability: 0,
+            valid: true
+          });
+        }
+      })
+      .catch(err => {
+        console.log(err);
+      })
+    });
+  }
 });
 
 console.log('>> WS server started on port', options.port);
+console.log('-- base path:', BASE_PATH)