I am stuck hereby as I am making an model which can classify art into three class whether it is painting , digital art , sculpture. I have a data set with 2k images each now I am stuck how to build a good model Can anyone tell me what is going wrong or what I am doing wrong . I checked metrices like precision and recall, they are upwards of 90% but accuracy lags behind (24% only). What could be the potential reasons? my data is clean I have two directories (training and validation) both have further subdirectories (corresponding to digitalart, painting, and sculpture with an equal number of samples) model = keras.Sequential([ # we are using the sequential api here it GOOD FOR simple model layers.Conv2D(32, (3, 3), activation='relu', input_shape=(256, 256, 3)), # this is first convolutional layer has 32 filters , using relu to introduce non-linearity # we define all these things in first layer later on all layers automatically inherit these props layers.MaxPooling2D(), # this reduces the dimension like image hieght and width layers.Conv2D(64, (3, 3), activation='relu'), # second layer layers.MaxPooling2D(), layers.Conv2D(128, (3, 3), activation='relu'), # third layers.MaxPooling2D(), layers.Flatten(), # converts 2d feature maps into one 1d to male fully connected layers layers.Dense(128, activation='relu'), # fully connected layer layers.Dense(3, activation='softmax') # 3 classes: painting, sculpture, digital art three neuron as three output ]) # Compile model model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy']) Continue reading...
