Why the validation accuracy is constant for all the pre-trained models while applying on medical datasets with .czi format?

Question

#import all the libraries
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
from sklearn.metrics import confusion_matrix,classification_report
import os
import sklearn.utils

path='/content/drive/My Drive/Drug/spheroid_imageset/untreated/training_set'

X=np.load(path+'/X.npy')
Y=np.load(path+'/Y.npy')

X=X.reshape(-1,224,224,1)

X,Y = sklearn.utils.shuffle(X,Y)


tf.keras.callbacks.EarlyStopping(
    # monitor="val_loss",
    # min_delta=0,
    patience=3,
    verbose=0,
    mode="auto",
    baseline=None,
    restore_best_weights=False,
)

image_size=224

base_model = tf.keras.applications.MobileNet(weights=None,include_top=False,input_shape=(image_size,image_size,1))
base_model.trainable = False


x = base_model.output
x = tf.keras.layers.GlobalAveragePooling2D()(x)
x = tf.keras.layers.Dropout(0.50)(x)
x = tf.keras.layers.Dense(16,activation='relu')(x)
x = tf.keras.layers.Dense(3,activation="softmax")(x)
model = tf.keras.Model(inputs=base_model.input, outputs=x)
model.compile(optimizer='adam',loss='sparse_categorical_crossentropy',metrics=['accuracy'])


history=model.fit(X,Y,epochs=50,validation_split=0.25,callbacks=[callback])

Answer 1

weights: One of None (random initialization), 'imagenet' (pre-training on ImageNet), or the path to the weights file to be loaded. Default to imagenet.

With weights=None, it is randomly initialized and
With base_model.trainable = False, it is not being trained.

So, basically your Conv. layer is not getting trained.