Model predict giving same results after fitting - tensorflow

Question

Hi I am a beginner in ML and Tensorflow so do pardon me for not understanding complex theories.
I was building an image classifier CNN as a form of practice. The model is trained using MobileNetv2 and its supposed to classify images of cats, dogs and pandas. After training my model (with decent accuracy of 92%), I tried using model.predict() to evaluate how it does with new images but I noticed that all my outputs were 1. This happens even if I used the same previous training data. By the way, I used 2700 (900 from each class) images for training and 300 for validation.
Here is my code
%tensorflow_version 2.x # this line is not required unless you are in a notebook
import tensorflow as tf
from tensorflow.keras import datasets, layers, models
from keras.preprocessing.image import load_img
from keras.preprocessing.image import img_to_array
from keras.preprocessing import image
from keras.preprocessing.image import ImageDataGenerator
from PIL import Image
import matplotlib.pyplot as plt
import numpy as np
import os
import PIL
import PIL.Image
import tensorflow_datasets as tfds
import pathlib
from google.colab import drive
drive.mount('/content/gdrive')
IMAGE_SIZE=[150,150]
train_path = "/content/gdrive/MyDrive/Colab Notebooks/Cat_Dog_Panda_CNN/train"
test_path = "/content/gdrive/MyDrive/Colab Notebooks/Cat_Dog_Panda_CNN/test"
IMAGE_SHAPE=[150,150,3]
base_model = tf.keras.applications.MobileNetV2(input_shape=IMAGE_SHAPE,
include_top=False,
weights='imagenet')
base_model.trainable = False
global_average_layer = tf.keras.layers.GlobalAveragePooling2D()
prediction_layer = tf.keras.layers.Dense(3)
model = tf.keras.Sequential([
base_model,
global_average_layer,
prediction_layer
])
model.summary()
base_learning_rate = 0.0001
model.compile(optimizer=tf.keras.optimizers.RMSprop(lr=base_learning_rate),
loss=tf.keras.losses.BinaryCrossentropy(from_logits=True),
metrics=['accuracy'])
# creates a data generator object that transforms images
train_datagen = ImageDataGenerator(
rescale=1./255,
rotation_range=40,
width_shift_range=0.2,
height_shift_range=0.2,
shear_range=0.2,
zoom_range=0.2,
horizontal_flip=True,
fill_mode='nearest')
test_datagen = ImageDataGenerator (rescale=1./255)
training_set = train_datagen.flow_from_directory(train_path, target_size=IMAGE_SIZE, batch_size=32, class_mode='categorical')
testing_set = test_datagen.flow_from_directory(test_path, target_size=IMAGE_SIZE, batch_size=32, class_mode='categorical')
model.fit(
training_set,
epochs=3,
validation_data=testing_set)
img = Image.open("/content/gdrive/MyDrive/Colab Notebooks/Cat_Dog_Panda_CNN/test/panda/panda_00094.jpg").convert('RGB').resize((150, 150), Image.ANTIALIAS)
img = np.array(img)
predictions = model.predict(img[None,:,:])
np.argmax(predictions[0])

Answer 1

in model.compile you have loss=tf.keras.losses.BinaryCrossentropy(from_logits=True). Since you have 3 classes you should have
loss=tf.keras.losses.CategoricalCrossentropy()
The Mobilenet model was trained with pixel values scaled within the range -1 to +1. So you should set rescale to
rescale=1/127.5-1
When you want to feed an image into model.predict you must perform the same preprocessing
of the image as you did for the training images namely rescale the image and resize the image to be the same size you used in training, namely (150,150).