Video Processing Using Deep Learning

Table of Contents

• Introduction
• How to Use
• Documentation
• License
• Authors
• References

Introduction

This repository utilizes deep learning models for video processing tasks, focusing on face detection, gender classification, and indoor/outdoor scene classification. It includes implementations using OpenCV and pretrained models such as MobileNetV2.

Face Detection and Gender Classification

For face detection and gender classification, OpenCV and deep learning models were employed. The models achieve significant accuracy in identifying faces and classifying genders in images and video streams.

checkout this notebook for demo Face detection and gender classification.ipynp

Indoor/Outdoor Scene Classification

Using a pretrained MobileNetV2 model on the SUN dataset, this project performs indoor/outdoor scene classification. The model was trained on 60,000 images, achieving high accuracy rates:

• Training Accuracy: 95.25%
• Validation Accuracy: 93.33%
• Testing Accuracy: 93.66%

Complete training details can be found in the Jupyter Notebook training classification model.ipynb

Video Processing using Jupyter Notebook

Here, 1000 videos have been tested for human detection, gender classification, indoor/outdoor classification, and also to provide video information such as resolution, frame rates, and duration. testing on 1000 videos.ipynb

Model weights are available (idod.weights.h5).

How to Use

To use this repository, follow these steps:

1. Clone the repository:

   git clone https://github.com/kishan9999/video-processing-using-deep-learning.git

2. Install the required dependencies:

   pip install -r requirements.txt

   Ensure the following packages are installed:

   • pandas==2.0.3
   • tensorflow==2.15.0
   • opencv-python==4.8.0.76
   • numpy==1.25.2
   • matplotlib==3.7.1
   • flask==2.2.5

3. Download All weights and place it in the weights folder

   • gender_deploy.prototxt
   • gender_net.caffemodel
   • opencv_face_detector_uint8.pb
   • opencv_face_detector.pbtxt

4. For video processing, use single_inference.py:

   python single_inference2.py --video_path

   OR replace path with your video file in python single_inference.py.

Video INFO API

run this code video_info.py to display information about the video file.

Documentation

For detailed documentation, refer to Document

License

This project is licensed under the MIT License.

Author

• Kishan Joshi

References

• Keras Models
• SUN dataset
• Age Gender Code Guide
• Age Gender Code Github
• Dataset used for action recognition: UCF101 Action Recognition