vid-prepper

Usage examples available at: https://github.com/jpetcon/vid-prepper/tree/main/examples

Package for preparing videos for deep learning models. Built on the giant shoulders of FFMPEG, NVIDIA DALI, OpenCV, Kornia, PySceneDetect, Torchvision and PyTorch.

This package attempts to bring some common video pre-processing methods together in an efficient way for both CPU and GPU.

Installation

Prerequisites

Before installing vid-prepper, you need to install the following external dependencies:

FFmpeg (Required)

FFmpeg is required for video processing operations.

Ubuntu/Debian:

sudo apt update
sudo apt install ffmpeg

macOS (with Homebrew):

brew install ffmpeg

Windows: Download from https://ffmpeg.org/download.html and add to PATH.

CUDA/NVDEC (Optional but Recommended)

For GPU acceleration, install CUDA toolkit and ensure your system supports NVDEC:

Ubuntu/Debian:

# Install CUDA toolkit
sudo apt install nvidia-cuda-toolkit

macOS: CUDA is not supported on macOS. The package will automatically fall back to CPU processing.

Windows: Download CUDA toolkit from https://developer.nvidia.com/cuda-downloads.

Package Installation

pip install vid-prepper

Usage

from vid_prepper import metadata, standardize

# Example usage
metadata_extractor = metadata.Metadata("video.mp4")
video_info = metadata_extractor.run()

standardizer = standardize.VideoStandardizer(
        size="224x224",
        fps=25,
        codec="h264",
        color="rgb",
        use_gpu=False  # Set to True if you have CUDA
    )

standardizer.standardize_video(video_input="video.mp4", output_path="video_standardized.mp4")

Features

• Video metadata extraction and validation
• Video standardization and preprocessing
• Object detection and scene analysis
• Video augmentation for deep learning
• Efficient tensor loading with GPU acceleration