yskaaks/soccer-video-preprocess

Given a full length soccer match video, this model gets rid of all frames that are not soccer game itself. Leaving only soccer playing frames ready for potential model training

Soccer Preprocessing Guide

This document outlines the setup and steps required to preprocess soccer match videos by removing unwanted segments, extracting relevant frames, and preparing data for model training

Environment Setup

1. Create a Conda Environment
   Start by creating a new Conda environment named soccer-preprocess with Python 3.11 installed.

   conda create -n soccer-preprocess python=3.11 -y
   

   Activate the environment:

   conda activate soccer-preprocess
   

2. Install PyTorch
   Install PyTorch, torchvision, torchaudio, and the corresponding CUDA toolkit version (e.g., for CUDA 11.7) using the commands below. Adjust the CUDA version based on your system's specifications.

   conda install pytorch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 pytorch-cuda=11.7 -c pytorch -c nvidia
   

   For different CUDA versions, refer to PyTorch's previous versions for guidance.

3. Additional Dependencies
   Install other necessary Python packages:

   pip install ultralytics supervision moviepy scikit-learn
   

Video Preprocessing

1. Trim Videos
   Remove pre-match programs and halftime advertisements by keeping only the match segments. For example, if the match is from 00:15:00 to 00:50:00 and 01:05:00 to 01:40:00, these segments are retained:

   python trim_videos.py /path/to/your/video.mp4 00:15:00-00:50:00 01:05:00-01:40:00
   

   Alternatively, you can use video editing software for this step. Store the trimmed videos in the initial-trimmed-videos directory.

2. Extract Frames
   To extract frames from the videos, adjust the VIDEO_DIR_PATH and IMAGES_DIR_PATH in get_video_frames.py and run:

   python get_video_frames.py
   

   This saves every 300th frame, producing approximately 2000 images (~1000 images per video).

3. Organize Frames
   Manually categorize the frames into relevant and irrelevant folders within a Data directory. Typically, use around 650 irrelevant frames and 950 relevant frames.

Data Preparation for Model Training

1. Prepare Training Data
   Update the paths in prepare_training_data.py and execute it to split the data into training, testing, and validation sets.

2. Model Training
   Navigate to the "Model Training" directory and initiate the training process:

   cd "Model Training"
   python train.py
   

Post-Training Processing

1. Remove Irrelevant Frames
   Inside the Remove Irrelevant Frames directory, adjust the paths in script.py and run it to segregate videos into relevant and irrelevant frames.

2. Remove Replays
   Manually record replay timestamps in text files (see Replay Timestamps folder for examples). Execute the following command to remove replays from videos:

   python remove_replays.py input_video_path output_video_path timestamps_file_path
   

This guide provides a comprehensive walkthrough for preprocessing soccer match videos, from environment setup to data preparation and model training. Follow each step carefully to ensure the successful completion of the process.