resiliency to invalid models config, fix #107

Author: previ

camera.py

@@ -87,9 +87,12 @@ def __init__(self):
         self._cnn_classifiers = {}
         cnn_model = config.Config.get().get("cnn_default_model", "")
         if cnn_model != "":
-            self._cnn_classifiers[cnn_model] = CNNManager.get_instance().load_model(cnn_model)
-            self._cnn_classifier_default = self._cnn_classifiers[cnn_model]
-            logging.info("loaded: " + cnn_model + " " + str(self._cnn_classifier_default))
+            try:
+                self._cnn_classifiers[cnn_model] = CNNManager.get_instance().load_model(cnn_model)
+                self._cnn_classifier_default = self._cnn_classifiers[cnn_model]
+                logging.info("loaded: " + cnn_model + " " + str(self._cnn_classifier_default))
+            except:
+                logging.warning("model not found: " + cnn_model)
 
         self._camera.grab_start()
         self._image_cv = self.get_image()
@@ -354,7 +357,7 @@ def find_ar_code(self):
         img = self.get_image()
         return img.find_ar_code()
 
-    def cnn_classify(self, model_name=None):
+    def cnn_classify(self, model_name=None, top_results=3):
         classifier = None
         if model_name:
             classifier = self._cnn_classifiers.get(model_name)
@@ -364,12 +367,17 @@ def cnn_classify(self, model_name=None):
         else:
             classifier = self._cnn_classifier_default
 
-        img = self.get_image()
-        classes = classifier.classify_image(img.mat())
+        classes = None
+        try:
+            img = self.get_image()
+            classes = classifier.classify_image(img.mat(), top_results=top_results) 
+        except:
+            logging.warning("classifier not available")
+            classes = [("None", 1.0)]
         return classes
 
     def find_class(self):
-        return self.cnn_classify()[0]
+        return self.cnn_classify(top_results=1)[0][0]
 
     def sleep(self, elapse):
         logging.debug("sleep: " + str(elapse))

cnn_classifier.py

@@ -110,24 +110,13 @@ def load_labels(self, label_file):
 
     def classify_image(self,
                        image_file_or_mat,
-                       input_height=128,
-                       input_width=128,
-                       input_mean=0,
-                       input_std=255):
+                       top_results=3):
         s_t = time.time()
         t = None
         if type(image_file_or_mat) == str:
-            t = self.read_tensor_from_image_file(file_name=image_file_or_mat,
-                                  input_height=input_height,
-                                  input_width=input_width,
-                                  input_mean=input_mean,
-                                  input_std=input_std)
+            t = self.read_tensor_from_image_file(file_name=image_file_or_mat)
         else:
-            t = self.read_tensor_from_image_mat(image_file_or_mat,
-                                  input_height=input_height,
-                                  input_width=input_width,
-                                  input_mean=input_mean,
-                                  input_std=input_std)
+            t = self.read_tensor_from_image_mat(image_file_or_mat)
 
         #logging.info( "time.norm: " + str(time.time() - s_t))
         s_t = time.time()
@@ -137,7 +126,9 @@ def classify_image(self,
 
         #logging.info( "time.cls: " + str(time.time() - s_t))
 
+        top_results = min(top_results, len(self._labels))
         results = np.squeeze(results)
-        result = results.argmax()
-        pairs = (self._labels[result], results[result])
+        results_idx = np.argpartition(results, -top_results)[-top_results:]
+        results_idx = np.flip(results_idx[np.argsort(results[results_idx])], axis=0)
+        pairs = [(self._labels[i], results[i]) for i in results_idx]
         return pairs

main.py

@@ -34,7 +34,7 @@
 
 # Logging configuration
 logger = logging.getLogger()
-logger.setLevel(logging.WARNING)
+logger.setLevel(logging.INFO)
 sh = logging.StreamHandler()
 fh = logging.handlers.RotatingFileHandler('./logs/coderbot.log', maxBytes=1000000, backupCount=5)
 formatter = logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s')