Using a TensorFlow Model - tensorflow

Question

https://www.tensorflow.org/tutorials/images/classification
TensorFlow has a beautiful tutorial on how to build an image classifier model to detect cats and dogs...
But they left out two crucial steps.
Step 1: How do you prepare an image to feed the just made model ?
Step 2: How do you feed the model ?
This is what I tried, with failing results.
new_array = cv2.imread('cat.jpg') "<--- CV2 Read Image"
dImg= new_array.reshape(1,150,150,3) "<-- Convert it to 4D input "
prediction = model.predict(dImg/255) "<---- Scale down by 255 ??? Idk im guessing "
print(str((prediction[0][0])) + " cat") "<-- Print the list of list prediction which rn gives unusable results"
Update
new_array = cv2.imread('dog.jpg')
dImg= new_array.reshape(1,150,150,3)
prediction = model.predict(dImg/255)
print(str((int(prediction[0][0]))) + ' dog')
print(prediction[0][0])
new_array = cv2.imread('dog-2.jpg')
dImg= new_array.reshape(1,150,150,3)
prediction = model.predict(dImg/255)
print(str((int(prediction[0][0]))) + ' dog')
print(prediction[0][0])
new_array = cv2.imread('dog-3.jpg')
dImg= new_array.reshape(1,150,150,3)
prediction = model.predict(dImg/255)
print(str((int(prediction[0][0]))) + ' dog')
print(prediction[0][0])
new_array = cv2.imread('dog-4.jpg')
dImg= new_array.reshape(1,150,150,3)
prediction = model.predict(dImg/255)
print(str((int(prediction[0][0]))) + ' dog')
print(prediction[0][0])
result
0 cat
0.5860402
-2 cat
-2.1347654
-1 cat
-1.380995
-4 cat
-4.0731945
1 dog
1.6571417
1 dog
1.759522
0 dog
-0.05260024
0 dog
-0.827193

Answer 1

To answer your questions and uncertainties:
"<---- Scale down by 255 ??? Idk im guessing "
This depends on how the model has been trained. If the images were divided to 255 before being fed to the model, then when you test you must divide to 255.
Testing on one image in Keras and TensorFlow means you need to add the batch_axis. In Keras and TensorFlow, one can only predict on batches of data.
Therefore, when loading an image (regardless of OpenCV and PIL):
image = cv2.imread(image_path) #note that opencv loads in BGR format
#If necessary convert to RGB(depends again on how the model was trained)
image = cv2.cvtColor(image,cv2.COLOR_BGR2RGB)
#If necessary divide by 255
image /= 255
image = np.expand_dims(image,axis=0) #or tf.expand_dims(image, axis=0)
model.predict(image)