Open raresct opened 7 years ago
The idea is to see whether training a new NN on top of existing features (or even the original image) works for the regression problem.
E.g. for resnet50
import numpy as np from keras.applications import ResNet50 from keras.layers import Dense, Flatten, Dropout, Input from keras.models import Model def custom_fc(): inputs = Input(shape=(1,1,2048)) x = Flatten()(inputs) x = Dropout(0.5, seed=1234)(x) x = Dense(256, activation='relu')(x) x = Dropout(0.1, seed=1234)(x) x = Dense(32, activation='relu')(x) outputs = Dense(1, activation='linear')(x) return Model(inputs=inputs, outputs=outputs) r50 = ResNet50(weights='imagenet',include_top=False,input_shape=(224,224,3)) x_train = np.random.random((10, 224, 224,3)) y_train = 100*np.random.random((10, 1)) x_train_r50 = r50.predict(x_train) print x_train_r50.shape model = custom_fc() model.compile(optimizer='rmsprop', loss='mse') model.fit(x_train_r50, y_train, epochs=100, batch_size=1) y_pred = model.predict(x_train_r50) print y_train print '*'*30 print y_pred
Also try binning target variable.
The idea is to see whether training a new NN on top of existing features (or even the original image) works for the regression problem.
E.g. for resnet50
Also try binning target variable.