Skip to content

Latest commit

 

History

History
68 lines (51 loc) · 2.13 KB

README.md

File metadata and controls

68 lines (51 loc) · 2.13 KB
Raw

vid23d Scallop

Introduction

vid23d_scallop is a Python application designed to transform 2D video frames into 3D stereo images using depth estimation and stereo pair generation. This project leverages deep learning models for depth perception and computer vision techniques to create enhanced visual experiences.

Features

  • Depth map estimation from 2D images.
  • Generation of stereo image pairs from single frames.
  • Video processing to extract frames and concatenate them back to video format.
  • Audio extraction and merging with processed video.

Prerequisites

Before you begin, ensure you have the following installed:

  • Python 3.8 or higher
  • OpenCV
  • Torch
  • tqdm
  • moviepy
  • matplotlib (optional, for debugging depth maps)

Installation

Clone the repository to your local machine:

git clone https://github.com/yourusername/vid23d_scallop.git
cd vid23d_scallop

Install the required Python libraries:

pip install -r requirements.txt

Usage

Configuring Your Device Edit conf.py to configure the processing device (CPU, CUDA, MPS):

import torch

device = 'cpu'
if torch.backends.mps.is_available():
    device = torch.device('mps')
if torch.cuda.is_available():
    device = torch.device('cuda')

print(f'Using device: {device}', flush=True)

Processing Videos

Use main.py to process videos through the command line. The script supports various operations including frame extraction, depth map creation, and stereo pair generation.

python main.py --input_video path/to/your/video.mp4 --output_dir path/to/output

Depth and Stereo Image Generation

img_depth.py and stereo.py are used internally to generate depth maps and stereo pairs, but can be used standalone for testing and development purposes.

Contributing

Contributions are welcome! Please fork the repository and submit a pull request with your new features or fixes.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgements

Our work uses code from MiDaS. We'd like to thank the authors for making this library available.