Skip to main content Link Menu Expand (external link) Document Search Copy Copied
e2evideo

Feature Extraction from Images Module Tutorial

This Python module is used for extracting features from images using a Convolutional Autoencoder (CAE). The code for the CAE is adapted from a tutorial on the Paperspace blog.

Dependencies

Before starting, make sure that you have the following libraries installed. If not, you can install them using pip:

pip install numpy torch torchvision scikit-learn

Main Function

The main function in this script performs either of the following tasks based on the argument mode:

If mode is ‘train’, it creates an instance of the Convolutional Autoencoder, trains it using the given training set, and saves the trained model.

If mode is not ‘train’, it loads a pre-trained model and uses it to generate and plot features for a set of test images.

Arguments

  • dataset_name: The name of the dataset. The default is ‘action_recognition’.
  • images_array: The path to the file containing the images array. The default is ‘./results/ucf10.npz’.
  • data_folder: The folder that contains the images. The default is '../data/images_ucf10/'.
  • labels_file: The path to the file containing the labels. The default is None.
  • no_classes: The number of classes in the dataset. The default is 10.
  • mode: The mode in which to run the script. The choices are ‘train’ and ‘test’. The default is ‘train’.
  • batch_size: The batch size to use during training. The default is 50.
  • epochs: The number of epochs to train the model for. The default is 50.

Example Usage

The typical usage from the command line would look like this:

python feature_extraction.py --dataset_name "action_recognition" --images_array "./results/ucf10.npz" --data_folder "../data/images_ucf10/" --labels_file "labels.txt" --no_classes 10 --mode "train" --batch_size 50 --epochs 50

This command will train the Convolutional Autoencoder using images in the ./results/ucf10.npz file and labels in the labels.txt file. It will run for 50 epochs with a batch size of 50.

If you want to use it inside another Python script, here is how you would do it:

import argparse
from feature_extraction import main, device

options = argparse.Namespace(
    dataset_name='action_recognition',
    images_array='./results/ucf10.npz',
    data_folder='../data/images_ucf10/',
    labels_file='labels.txt',
    no_classes=10,
    mode='train',
    batch_size=50,
    epochs=50
)

main(options)

* This tutorial has been made with the help of ChatGPT 4.0.