Home
Deep Learning For Vision

Study Pretrained Deep Neural Network Models for Images

Aim

To study the usage of pretrained deep neural network models for image classification, demonstrating feature extraction and prediction using transfer learning in Python (with Keras and TensorFlow).

Algorithm

  1. Load a Pretrained Model (e.g., VGG16) with pre-learned weights on ImageNet.

  2. Prepare Input Image: Load and resize the image to match the network input size, apply required preprocessing.

  3. Predict: Pass the image through the pretrained model to obtain predictions.

  4. Interpret Results: Decode the model’s output to obtain human-readable class labels.

Program



open cmd or terminal
and then type "pip install tensorflow" to install the required packages
from tensorflow.keras.applications import VGG16
from tensorflow.keras.preprocessing import image
from tensorflow.keras.applications.vgg16 import preprocess_input, decode_predictions
import numpy as np

# 1. Load pretrained VGG16 model
model = VGG16(weights='imagenet')

# 2. Load and preprocess sample image
img_path = 'elephant.jpg'  # Replace with your image
img = image.load_img(img_path, target_size=(224, 224))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = preprocess_input(x)

# 3. Predict using the model
preds = model.predict(x)

# 4. Output top 3 predictions
print('Predicted:', decode_predictions(preds, top=3)[0])

Output (Sample: Elephant Image)



Predicted: [('n02504458', 'African_elephant', 0.86), ('n01871265', 'tusker', 0.09), ('n02504013', 'Indian_elephant', 0.05)]

Result

  • The pretrained model can classify images with high accuracy, revealing top likely objects detected.

  • This exercise shows how transfer learning enables leveraging large-scale image features for new tasks without retraining networks from scratch.

  • The same approach supports feature extraction: intermediate model outputs can be used for clustering, retrieval, or further fine-tuning.