Why training results are different with pred in keras50

Question

I'm a newbie in Machine Learning. I want to build a keras model which will be used for image recognition. I am currently using the resnet50 model at:

PIL.Image.open(str(rock[0]))
img_height, img_width = 224,224
batch_size = 32
trains_ds = tf.keras.preprocessing.image_dataset_from_directory(
  data_dir,
  validation_split = 0.2,
  subset = "training",
  seed = 42,
  label_mode = 'categorical',
  image_size = (img_height, img_width),
  batch_size = batch_size)
val_ds = tf.keras.preprocessing.image_dataset_from_directory(
    data_dir,
    validation_split=0.2,
    subset="validation",
    seed=42,
    label_mode = 'categorical',
    image_size=(img_height, img_width),
    batch_size=batch_size)

class_names = trains_ds.class_names

resnet_model = Sequential()
pretrained_model = tf.keras.applications.ResNet50(include_top=False, 
                                                  input_shape=(224,224,3),
                                                  pooling='avg', 
                                                  classes = 20, 
                                                  weights = 'imagenet')

for layer in pretrained_model.layers: 
    layer.trainable=False
resnet_model.add(pretrained_model)
resnet_model.add(Flatten())
resnet_model.add(Dense(512, activation='relu'))
resnet_model.add(Dense(20,activation='softmax'))
resnet_model.summary()
resnet_model.compile(optimizer=Adam(learning_rate=0.001),loss='categorical_crossentropy',metrics=['accuracy'])
epochs = 5
history= resnet_model.fit(
    trains_ds,
    validation_data=val_ds,
    epochs=epochs)
image = cv2.imread(str(rock[0]))
image_resized = cv2.resize(image,(img_height,img_width))
print(img_height)
print(img_width)
image=np.expand_dims(image_resized,axis=0)
print(image.shape)
pred=resnet_model.predict(image)
print(pred)
output_class=class_names[np.argmax(pred)]
print("The predicted class is", output_class)

There are 20 classes. Train the model, the results looks great. The train accuracy and the val accuracy are almost 100%. But when I choose some train file from the train_ds and some files from the val_ds, the predicit results are bad. I don't know why. Is this model overfitting? or?