add negatives mining strategies

.gitignore

@@ -113,3 +113,4 @@ encodings/
 weights/
 plots/
 sub.csv
+core

configs/road_signs.yml

@@ -8,6 +8,7 @@ optimizer : 'radam'
 learning_rate : 0.0001
 project_name : 'road_signs/'
 freeze_backbone : True
+embeddings_normalization: True
 
 #paths
 dataset_path : '/home/rauf/datasets/road_signs/road_signs_separated/'

configs/road_signs_mobilenetv2.yml

@@ -8,6 +8,7 @@ optimizer : 'radam'
 learning_rate : 0.0001
 project_name : 'road_signs/'
 freeze_backbone : True
+embeddings_normalization: True
 
 #paths
 dataset_path : '/home/rauf/datasets/road_signs/road_signs_separated/'

configs/road_signs_resnet18.yml

@@ -8,6 +8,7 @@ optimizer : 'radam'
 learning_rate : 0.0001
 project_name : 'road_signs/'
 freeze_backbone : True
+embeddings_normalization: True
 
 #paths
 dataset_path : '/home/rauf/datasets/road_signs/road_signs_separated/'

configs/road_signs_resnet50v2.yml

@@ -8,6 +8,7 @@ optimizer : 'radam'
 learning_rate : 0.0001
 project_name : 'road_signs/'
 freeze_backbone : True
+embeddings_normalization: True
 
 #paths
 dataset_path : '/home/rauf/datasets/road_signs/road_signs_separated/'

configs/road_signs_resnet50v2_merged_dataset.yml

@@ -8,6 +8,7 @@ optimizer : 'radam'
 learning_rate : 0.0001
 project_name : 'road_signs/'
 freeze_backbone : True
+embeddings_normalization: True
 
 #paths
 dataset_path : '/home/rauf/datasets/road_signs_merged/'

configs/road_signs_resnext50.yml

@@ -8,6 +8,7 @@ optimizer : 'radam'
 learning_rate : 0.0001
 project_name : 'road_signs/'
 freeze_backbone : True
+embeddings_normalization: True
 
 #paths
 dataset_path : '/home/rauf/datasets/road_signs/road_signs_separated/'

configs/road_signs_resnext50_merged_dataset.yml

@@ -7,7 +7,8 @@ backbone_weights : 'imagenet'
 optimizer : 'radam'
 learning_rate : 0.0001
 project_name : 'road_signs/'
-freeze_backbone : False
+freeze_backbone : True
+embeddings_normalization: True
 
 #paths
 dataset_path : '/home/rauf/datasets/road_signs_merged/'

configs/road_signs_simple2_merged_dataset.yml

@@ -8,6 +8,7 @@ optimizer : 'radam'
 learning_rate : 0.0001
 project_name : 'road_signs/'
 freeze_backbone : True
+embeddings_normalization: True
 
 #paths
 dataset_path : '/home/rauf/datasets/road_signs_merged/'

siamese_net/backbones.py

@@ -3,79 +3,94 @@
 from classification_models import Classifiers
 from keras.models import Model
 from keras.regularizers import l2
+import keras.backend as K
 
-def get_backbone(input_shape,encodings_len=4096,backbone_type='simple',backbone_weights='imagenet',freeze_backbone=False):
-        if backbone_type == 'simple':
-            input_image = Input(input_shape)
-            x = Conv2D(64, (10, 10), activation='relu',
-                       kernel_regularizer=l2(2e-4))(input_image)
-            x = MaxPool2D()(x)
-            x = Conv2D(128, (7, 7), activation='relu',
-                       kernel_regularizer=l2(2e-4))(x)
-            x = MaxPool2D()(x)
-            x = Conv2D(128, (4, 4), activation='relu',
-                       kernel_regularizer=l2(2e-4))(x)
-            x = MaxPool2D()(x)
-            x = Conv2D(256, (4, 4), activation='relu',
-                       kernel_regularizer=l2(2e-4))(x)
-            x = Flatten()(x)
-            encoded_output = Dense(encodings_len, activation='sigmoid',
-                                   kernel_regularizer=l2(1e-3))(x)
-            base_model = Model(
-                inputs=[input_image], outputs=[encoded_output])
-        elif backbone_type == 'simple2':
-            input_image = Input(input_shape)
-            x = Conv2D(32, kernel_size=3, activation='relu',
-                       kernel_regularizer=l2(2e-4))(input_image)
-            x = BatchNormalization()(x)
-            x = Conv2D(32, kernel_size=3, activation='relu',
-                       kernel_regularizer=l2(2e-4))(x)
-            x = BatchNormalization()(x)
-            x = Conv2D(32, kernel_size=5, strides=2, padding='same', activation='relu',
-                       kernel_regularizer=l2(2e-4))(x)
-            x = BatchNormalization()(x)
-            x = Dropout(0.4)(x)
 
-            x = Conv2D(64, kernel_size=3, activation='relu',
-                       kernel_regularizer=l2(2e-4))(x)
-            x = BatchNormalization()(x)
-            x = Conv2D(64, kernel_size=3, activation='relu',
-                       kernel_regularizer=l2(2e-4))(x)
-            x = BatchNormalization()(x)
-            x = Conv2D(64, kernel_size=5, strides=2, padding='same', activation='relu',
-                       kernel_regularizer=l2(2e-4))(x)
-            x = BatchNormalization()(x)
-            x = Dropout(0.4)(x)
+def get_backbone(input_shape,
+                 encodings_len=4096,
+                 backbone_type='simple',
+                 embeddings_normalization=True,
+                 backbone_weights='imagenet',
+                 freeze_backbone=False):
+    if backbone_type == 'simple':
+        input_image = Input(input_shape)
+        x = Conv2D(64, (10, 10), activation='relu',
+                   kernel_regularizer=l2(2e-4))(input_image)
+        x = MaxPool2D()(x)
+        x = Conv2D(128, (7, 7), activation='relu',
+                   kernel_regularizer=l2(2e-4))(x)
+        x = MaxPool2D()(x)
+        x = Conv2D(128, (4, 4), activation='relu',
+                   kernel_regularizer=l2(2e-4))(x)
+        x = MaxPool2D()(x)
+        x = Conv2D(256, (4, 4), activation='relu',
+                   kernel_regularizer=l2(2e-4))(x)
+        x = Flatten()(x)
+        encoded_output = Dense(encodings_len, activation='relu',
+                               kernel_regularizer=l2(1e-3))(x)
+        if embeddings_normalization:
+            encoded_output = Lambda(lambda x: K.l2_normalize(
+                x, axis=1), name='l2_norm')(encoded_output)
+        base_model = Model(
+            inputs=[input_image], outputs=[encoded_output])
+    elif backbone_type == 'simple2':
+        input_image = Input(input_shape)
+        x = Conv2D(32, kernel_size=3, activation='relu',
+                   kernel_regularizer=l2(2e-4))(input_image)
+        x = BatchNormalization()(x)
+        x = Conv2D(32, kernel_size=3, activation='relu',
+                   kernel_regularizer=l2(2e-4))(x)
+        x = BatchNormalization()(x)
+        x = Conv2D(32, kernel_size=5, strides=2, padding='same', activation='relu',
+                   kernel_regularizer=l2(2e-4))(x)
+        x = BatchNormalization()(x)
+        x = Dropout(0.4)(x)
 
-            x = Conv2D(128, kernel_size=4, activation='relu',
-                       kernel_regularizer=l2(2e-4))(x)
-            x = BatchNormalization()(x)
-            x = Flatten()(x)
-            x = Dense(512, activation="relu")(x)
-            x = Dropout(0.5)(x)
-            encoded_output = Dense(encodings_len, activation='sigmoid',
-                                   kernel_regularizer=l2(1e-3))(x)
-            base_model = Model(
-                inputs=[input_image], outputs=[encoded_output])
-        else:
-            classifier, preprocess_input = Classifiers.get(backbone_type)
-            backbone_model = classifier(input_shape=input_shape, 
-                                        weights=backbone_weights, 
-                                        include_top=False)
+        x = Conv2D(64, kernel_size=3, activation='relu',
+                   kernel_regularizer=l2(2e-4))(x)
+        x = BatchNormalization()(x)
+        x = Conv2D(64, kernel_size=3, activation='relu',
+                   kernel_regularizer=l2(2e-4))(x)
+        x = BatchNormalization()(x)
+        x = Conv2D(64, kernel_size=5, strides=2, padding='same', activation='relu',
+                   kernel_regularizer=l2(2e-4))(x)
+        x = BatchNormalization()(x)
+        x = Dropout(0.4)(x)
 
-            if freeze_backbone:
-                for layer in backbone_model.layers[:-2]:
-                    layer.trainable = False
+        x = Conv2D(128, kernel_size=4, activation='relu',
+                   kernel_regularizer=l2(2e-4))(x)
+        x = BatchNormalization()(x)
+        x = Flatten()(x)
+        x = Dense(512, activation="relu")(x)
+        x = Dropout(0.5)(x)
+        encoded_output = Dense(encodings_len, activation='relu',
+                               kernel_regularizer=l2(1e-3))(x)
+        if embeddings_normalization:
+            encoded_output = Lambda(lambda x: K.l2_normalize(
+                x, axis=1), name='l2_norm')(encoded_output)
+        base_model = Model(
+            inputs=[input_image], outputs=[encoded_output])
+    else:
+        classifier, preprocess_input = Classifiers.get(backbone_type)
+        backbone_model = classifier(input_shape=input_shape,
+                                    weights=backbone_weights,
+                                    include_top=False)
 
-            after_backbone = backbone_model.output
-            x = Flatten()(after_backbone)
-            # x = Dense(512, activation="relu")(x)
-            # x = Dropout(0.5)(x)
-            # x = Dense(512, activation="relu")(x)
-            # x = Dropout(0.5)(x)
-            encoded_output = Dense(encodings_len, activation="relu")(x)
+        if freeze_backbone:
+            for layer in backbone_model.layers[:-2]:
+                layer.trainable = False
 
-            base_model = Model(
-                inputs=[backbone_model.input], outputs=[encoded_output])
+        after_backbone = backbone_model.output
+        x = Flatten()(after_backbone)
+        # x = Dense(512, activation="relu")(x)
+        # x = Dropout(0.5)(x)
+        # x = Dense(512, activation="relu")(x)
+        # x = Dropout(0.5)(x)
+        encoded_output = Dense(encodings_len, activation="relu")(x)
+        if embeddings_normalization:
+            encoded_output = Lambda(lambda x: K.l2_normalize(
+                x, axis=1), name='l2_norm')(encoded_output)
+        base_model = Model(
+            inputs=[backbone_model.input], outputs=[encoded_output])
 
-        return base_model
+    return base_model