# ResNet50 Model is not learning with transfer learning in keras

I am trying to perform transfer learning on ResNet50 model pretrained on Imagenet weights for PASCAL VOC 2012 dataset. As it is a multi label dataset, I am using sigmoid activation function in the final layer and binary_crossentropy loss. The metrics are precision,recall and accuracy. Below is the code I used to build the model for 20 classes (PASCAL VOC has 20 classes).

img_height,img_width = 128,128
num_classes = 20
#If imagenet weights are being loaded,
#input must have a static square shape (one of (128, 128), (160, 160), (192, 192), or (224, 224))
base_model = applications.resnet50.ResNet50(weights= 'imagenet', include_top=False, input_shape= (img_height,img_width,3))
x = base_model.output
x = GlobalAveragePooling2D()(x)
#x = Dropout(0.7)(x)
predictions = Dense(num_classes, activation= 'sigmoid')(x)
model = Model(inputs = base_model.input, outputs = predictions)
for layer in model.layers[-2:]:
layer.trainable=True
for layer in model.layers[:-3]:
layer.trainable=False

#print(model.summary())

X_train, X_test, Y_train, Y_test = train_test_split(x_train, y, random_state=42, test_size=0.2)
savingcheckpoint = ModelCheckpoint('ResnetTL.h5',monitor='val_loss',verbose=1,save_best_only=True,mode='min')
earlystopcheckpoint = EarlyStopping(monitor='val_loss',patience=10,verbose=1,mode='min',restore_best_weights=True)
model.fit(X_train, Y_train, epochs=epochs, validation_data=(X_test,Y_test), batch_size=batch_size,callbacks=[savingcheckpoint,earlystopcheckpoint],shuffle=True)
model.save_weights('ResnetTLweights.h5')


It ran for 35 epochs until earlystopping and the metrics are as follows (without Dropout layer):

loss: 0.1195 - accuracy: 0.9551 - precision_m: 0.8200 - recall_m: 0.5420 - val_loss: 0.3535 - val_accuracy: 0.8358 - val_precision_m: 0.0583 - val_recall_m: 0.0757


Even with Dropout layer, I don't see much difference.

loss: 0.1584 - accuracy: 0.9428 - precision_m: 0.7212 - recall_m: 0.4333 - val_loss: 0.3508 - val_accuracy: 0.8783 - val_precision_m: 0.0595 - val_recall_m: 0.0403


With dropout, for a few epochs, the model is reaching to a validation precision and accuracy of 0.2 but not above that.

I see that precision and recall of validation set is pretty low compared to training set with and without dropout layer. How should I interpret this? Does this mean the model is overfitting. If so, what should I do? As of now the model predictions are quite random (totally incorrect). The dataset size is 11000 images.

• your high validation accuracy and low validation precision show that your validation data is skewed. Please add validation and test data class distribution. – shivam shah Oct 15 '19 at 9:17