test.py

import tensorflow as tf
import numpy as np
import h5py
import cv2
import pickle

# write list to binary file
def write_list(a_list):
    # store list in binary file so 'wb' mode
    with open('/home/bob/fungi-id-ai/model/fungi_model_texts_filtered_20.bin', 'wb') as fp:
        pickle.dump(a_list, fp)
        print('Done writing list into a binary file')

# Read list to memory
def read_list():
    # for reading also binary mode is important
    with open('/home/bob/fungi-id-ai/model/fungi_model_texts_filtered_20.bin', 'rb') as fp:
        n_list = pickle.load(fp)
        return n_list

# Open the HDF5 dataset file
h5f = h5py.File('/home/bob/fungi-id-ai/model/fungi_model_filtered_20.h5', 'r')

# Load the data from the HDF5 file
# model = tf.keras.models.load_model('/home/bob/fungi-id-ai/model/fungi_pretrained_model_rgb_128_full_gpu.h5')
model = tf.keras.models.load_model('/home/bob/fungi-id-ai/model/fungi_pretrained_model_filtered_20.h5')

# Load the image
image = cv2.imread('/home/bob/fungi-id-ai/data/image_4.jpg', cv2.IMREAD_COLOR)
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)

# Crop the image
h, w, c = image.shape
start_x = 0
start_y = 0
end_x = w
end_y = h
if w > h:
    start_x = (w - h) // 2
    end_x = w - start_x
if h > w:
    start_y = (h - w) // 2
    end_y = h - start_y

cropped_image = image[start_y:end_y, start_x:end_x]

# Resize the images to 128x128
processed_image = cv2.resize(cropped_image, (128, 128), interpolation=cv2.INTER_AREA)

# Preprocess the image using OpenCV functions
processed_image = processed_image.astype('float32') / 255.0

# Predict the class of the image
prediction = model.predict(np.array([processed_image]))

# Print the prediction
print(prediction[0])

predicted_class = np.argmax(prediction[0])

# Get the label of the predicted class
class_labels = h5f['labels'][:]
unique_classes = np.unique(class_labels)
predicted_label = read_list()[predicted_class]

print("The predicted label for the new image is:", predicted_label)

# Calculate accuracy
accuracy = tf.keras.metrics.Accuracy()
accuracy.update_state(class_labels[predicted_class], class_labels[np.argmax(prediction[0])])
print("Accuracy: {:.2f}%".format(prediction[0][predicted_class] * 100))

# The predicted label for the new image is: 131
# Close the HDF5 file
h5f.close()