Classify Images with Pre-Trained Model using Keras

Classify Images with Pre-Trained Model using Keras

Computer Vision models have come a long way - and you can leverage existing models, pre-trained on large corpora of data, and just plug them into your prediction pipeline.

While fine-tuning a network is the best way to go - importing an existing model and running predictions from the get-go is a valid use-case in early stages of prototyping or just for the sake of trying a model out. For a detailed in-depth guide on fine-tuning via Transfer Learning - read our free lesson on "Transfer Learning for Computer Vision with Keras"!

That being said - the main variable is how you load an image in, the most notable ways being using OpenCV, PIL and the Keras preprocessing helper methods.

OpenCV

OpenCV is to Computer Vision, what Scikit-Learn is to Machine Learning. It's a popular, well-rounded library, widely used in the industry, with a slightly steeper learning curve than image processing libraries like PIL.

We'll create a proprietary helper function to obtain and convert an image via a URL into a NumPy array and then classify it using a pre-trained model from keras.applications:

from tensorflow import keras
# preprocess_input is a pass-through function for EffNets
from keras.applications.efficientnet import preprocess_input, decode_predictions

import urllib
import cv2
import numpy as np

def url_to_array(url):
    req = urllib.request.urlopen(url)
    arr = np.array(bytearray(req.read()), dtype=np.int8)
    arr = cv2.imdecode(arr, -1)
    arr = cv2.cvtColor(arr, cv2.COLOR_BGR2RGB)
    arr = cv2.resize(arr, (224, 224))
    return arr

url = 'https://upload.wikimedia.org/wikipedia/commons/0/02/Black_bear_large.jpg'
img = url_to_array(url)
img_batch = np.expand_dims(img, 0)

effnet = keras.applications.EfficientNetV2B0(weights='imagenet', include_top=True)
pred = effnet.predict(img_batch)
print(decode_predictions(pred))
# [[('n02133161', 'American_black_bear', 0.8694465), ('n02134418', 'sloth_bear', 0.03677345), ('n02132136', 'brown_bear', 0.020618025), ('n02134084', 'ice_bear', 0.0011013364), ('n02128925', 'jaguar', 0.00081196934)]]

Let's visualize the image with the prediction:

plt.imshow(img)
plt.title(f'Class: {decode_predictions(pred)[0][0][1]}\nConfidence: {decode_predictions(pred)[0][0][2]*100}%')
plt.show()

Keras Preprocessing

Keras offers preprocessing helper functions and classes which allows us to load and prepare images as NumPy arrays:

from tensorflow import keras
# preprocess_input is a pass-through function for EffNets
from keras.applications.efficientnet import preprocess_input, decode_predictions
from tensorflow.keras.preprocessing import image

import urllib.request
import matplotlib.pyplot as plt
import numpy as np

# Public domain image
url = 'https://upload.wikimedia.org/wikipedia/commons/0/02/Black_bear_large.jpg'
urllib.request.urlretrieve(url, 'bear.jpg')

# Load image and resize (doesn't keep aspect ratio)
img = image.load_img('bear.jpg', target_size=(224, 224))
# Turn to array of shape (224, 224, 3)
img = image.img_to_array(img)
# Expand array into (1, 224, 224, 3)
img_batch = np.expand_dims(img, 0)
# Preprocess for models that have specific preprocess_input() function
# img_preprocessed = preprocess_input(img)

# Load model and run prediction
effnet = keras.applications.EfficientNetV2B0(weights='imagenet', include_top=True)
pred = effnet.predict(img_batch)
print(decode_predictions(pred))
# [[('n02133161', 'American_black_bear', 0.87167954), ('n02134418', 'sloth_bear', 0.035227217), ('n02132136', 'brown_bear', 0.021057261), ('n02134084', 'ice_bear', 0.0011267527), ('n02128925', 'jaguar', 0.0007899802)]]

Get free courses, guided projects, and more

No spam ever. Unsubscribe anytime. Read our Privacy Policy.

Let's plot it:

plt.imshow(img.astype('int'))
plt.title(f'Class: {decode_predictions(pred)[0][0][1]}\nConfidence: {decode_predictions(pred)[0][0][2]*100}%')
plt.show()

PIL

Finally, PIL is a generally popular image processing library and can naturally be used to load images as NumPy arrays:

from tensorflow import keras
# preprocess_input is a pass-through function for EffNets
from keras.applications.efficientnet import preprocess_input, decode_predictions
from tensorflow.keras.preprocessing import image

import PIL

import urllib.request
import matplotlib.pyplot as plt
import numpy as np

# Public domain image
url = 'https://upload.wikimedia.org/wikipedia/commons/0/02/Black_bear_large.jpg'
img = PIL.Image.open(urllib.request.urlopen(url))
img = img.resize((224, 224))
img_batch = np.expand_dims(img, 0)

# Load model and run prediction
effnet = keras.applications.EfficientNetV2B0(weights='imagenet', include_top=True)
pred = effnet.predict(img_batch)
print(decode_predictions(pred))
# [[('n02133161', 'American_black_bear', 0.82402), ('n02134418', 'sloth_bear', 0.07482512), ('n02132136', 'brown_bear', 0.01921546), ('n02483708', 'siamang', 0.001174309), ('n02134084', 'ice_bear', 0.00081004837)]]

Let's plot it:

plt.imshow(img)
plt.title(f'Class: {decode_predictions(pred)[0][0][1]}\nConfidence: {decode_predictions(pred)[0][0][2]*100}%')
plt.show()
Was this helpful?
David LandupAuthor

Entrepreneur, Software and Machine Learning Engineer, with a deep fascination towards the application of Computation and Deep Learning in Life Sciences (Bioinformatics, Drug Discovery, Genomics), Neuroscience (Computational Neuroscience), robotics and BCIs.

Great passion for accessible education and promotion of reason, science, humanism, and progress.

Project

DeepLabV3+ Semantic Segmentation with Keras

# python# machine learning# tensorflow# computer vision

Semantic segmentation is the process of segmenting an image into classes - effectively, performing pixel-level classification. Color edges don't necessarily have to be the boundaries...

David Landup
David Landup
Details
Project

Building Your First Convolutional Neural Network With Keras

# python# artificial intelligence# machine learning# tensorflow

Most resources start with pristine datasets, start at importing and finish at validation. There's much more to know. Why was a class predicted? Where was...

David Landup
David Landup
Details

© 2013-2022 Stack Abuse. All rights reserved.

DisclosurePrivacyTerms