This document provides a step-by-step guide to install the necessary libraries, download the required models and config files, and execute the method for the SurgVU2024 Category-2 submission. You will also find instructions for exporting the model in a container for submission.

1. Clone Repository
2. Install Libraries
3. Download Model and Config
4. Running Method
5. Export container

To begin, you need to clone the repository that contains the codebase for the project.

git clone https://github.com/quzanh1130/SurgVU2024-Category-2-Submission.git
cd SurgVU2024-Category-2-Submission

Our code is based on mmaction2. To run our code, you must install mmaction2.

On Ubuntu or Debian

sudo apt update && sudo apt install ffmpeg

On Arch Linux

sudo pacman -S ffmpeg

On MacOS using Homebrew (https://brew.sh/)

brew install ffmpeg

On Windows using Chocolatey (https://chocolatey.org/)

choco install ffmpeg

On Windows using Scoop (https://scoop.sh/)

scoop install ffmpeg

Using Anaconda

conda create --name openmmlab python=3.8 -y
conda activate openmmlab

# Install pytorch at: https://pytorch.org/get-started/locally/
conda install pytorch torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia 
pip install -U openmim
mim install mmengine
mim install mmcv==2.1.0

cd mmaction2
pip install -v -e .

First you need to create folder to store model

cd mmaction2
mkdir work_dirs
cd work_dirs
mkdir slowfast

If the model file is missing at mmaction2/work_dirs/slowfast/best_model.pth, you can download it from the following Google Drive link:

• Download Model

If the config file is missing at mmaction2/configs/recognition/slowfast/slowfast_r101_8xb8-8x8x1-256e_rgb_1fps_video_final.py, you can download it from the following Google Drive link:

• Download Config

To run the model and test your input video, execute the following steps:

1. Navigate to the directory where the process_testing.py file is located.

2. Run the script using the following command:

python process_testing.py

3. After successful execution, the results will be saved in a JSON file at:

 mmaction2/output/surgical-step-classification.json

To use different configurations or model checkpoints, modify the process_testing.py file:

#Config and checkpoint
config_path = 'configs/recognition/slowfast/slowfast_r101_8xb8-8x8x1-256e_rgb_1fps_video_final.py'
checkpoint_path = 'work_dirs/slowfast/best_model.pth'

To test a different input video, modify the process_testing.py file:

if __name__ == "__main__":
    detector = SurgVUClassify()
    video_file = detector.input_path / "vid_1_short.mp4" # Change your test video
    detector.predict(str(video_file))

To prepare your submission, you will need to export the containerized application using Docker. If you don't have Docker installed, follow the links below for installation instructions:

1. Install Docker on Ubuntu

2. Install Docker on Window

Once Docker is installed, navigate to the mmaction2 directory and run the following command to export the container:

cd mmaction2
bash export.sh

After successful execution, a tarball of the container will be created at:

mmaction2/surgtoolloc_det.tar.gz

This file is your submission container.