predicter.py

import tensorflow as tf
import numpy as np
import os
from glob import glob


def base(path):
    return os.path.basename(path)


physical_devices = tf.config.list_physical_devices('GPU')
tf.config.experimental.set_memory_growth(physical_devices[0], True)

image_size = (100, 100)

class_names = glob("/home/wordp/fruits-360/Test/*") #Switch with a generator?
class_names = list(sorted(map(base, class_names)))

model = tf.keras.models.load_model('models')

img = tf.keras.preprocessing.image.load_img(
    r'/home/wordp/Fruit-Recognizer-master/p5.jpg', target_size=image_size
)

img_array = tf.keras.preprocessing.image.img_to_array(img)
img_array = tf.expand_dims(img_array, 0)

predictions = model.predict(img_array)

top_k_values, top_k_indices = tf.nn.top_k(predictions, k=5)
top_k_indices, top_k_values = (top_k_indices.numpy()[0], top_k_values.numpy()[0])

for idx in range(len(top_k_values)):
    print(
        "This image most likely belongs to {} with a {:.2f} percent confidence."
            .format(class_names[top_k_indices[idx]], 100 * np.max(top_k_values[idx]))
    )