Zhuifeng414 / AISG-Online-Safety-Challenge-Submission-Guide

Submission Guide + Discussion Board for AI Singapore Online Safety Prize Challenge

AI Singapore Online Safety Prize Challenge Submission Guide

Participants must submit a compressed Docker container in the tar.gz format via the challenge platform. This repository serves as a step-by-step guide to help participants create a valid submission for the Online Safety Prize Challenge.

While the proper term for the Docker generated artefacts is "Docker images", we will use the term "Docker container" instead to disambiguate it from the [computer] images that serve as input to the challenge.

Getting Started

We are using Docker for this challenge so that participants can choose their preferred programming languages and open source dependencies to create the best performing detection models.

To build and run GPU accelerated Docker containers, please install the NVIDIA Container Toolkit in your development environment.

Technical Details

Hardware/Software Specifications

All participants' compressed Docker containers will be executed on virtual machines with the following resource allocation:

vCPU	Mem (GB)	GPU	tmpfs (GiB)
6	60	V100 16GB VRAM	5

This will be reflected in the docker run command options. Participants may specify different settings for their own testing purposes, but these will not be reflected in the official run-time environment for scoring.

The general software specification

• Ubuntu 22.04
• NVIDIA Driver Version: 535
  • CUDA 11.x-12.x
  • Check for CUDA - NVIDIA Driver Compatibility

Submission Specification Guidelines

This section will cover the following important guidelines on building your solution for submission:

1. A brief overview of the dataset;
2. The required input format for your submitted Docker container and the output format from it;
3. The maximum resources of a Docker container for each submission;
4. The performance metrics for this challenge;
5. Instructions on how to run this repository and create your own submission.

Dataset, Input, Output

Dataset:

The dataset comprises about 1700 images in PNG format with file extension .png. The directory containing the dataset will be mounted to your Docker container as read-only at the mount point of /images.

Input to Docker Container:

Your solution must use stdin, where each line from stdin corresponds to a file path to a single image. This simulates a real-world situation where your solution is called on a per-image basis. An example of a line of input from stdin might look like /images/9ad157be-f32f-4770-b409-8c4650478f5b.png .

Further details on how this is done for a Python-based Docker solution can be found in Usage of sample submission and Creating your own submission.

Output (stdout, stderr) to Container:

Solution output: stdout

Your solution must use stdout to output the result of your analysis for each line of input from stdin.

The output format per line of input of stdin must include:

• The probability that the meme is harmful with 4 decimal places of precision. We will refer to this as proba.
• An integer, where 1 is output if the meme is deemed harmful, and 0 if it is benign. We will refer to this as label.

Both predictions must be separated by a single tab character (\t), and terminated with a single new line character (\n).

Examples of implementation for the formatting of one line of output in Python can be seen below.

# Example 1:
output = f"{proba:.4f}\t{label}\n"

# Example 2 (more C-like)
output = "%.4f\t%d\n" % (proba, label)

sys.stdout.write(output)

Below shows an example of how stdout will look like after having processed (in this case) 5 lines of input from stdin:

0.8232	1
0.7665	1
0.3241	0
0.1015	0
0.9511	1

Further details on how this is done for a Python-based Docker solution can be found in Usage of sample submission and Creating your own submission.

Remember that there has to be one output format conforming line in stdout for every input line from stdin. Please do not attempt to skip any.

Failure to do so may result in inaccurate scoring of your results.

Logging output: stderr

Your solution must use stderr for the writing of any logs to assist you in determining any programming errors within your solution. Logs have an implied file size limit to prevent abuse. Failure to keep within this limit through excessive logging will result in an error in your solution.

Further details on how this is done for a Python-based Docker solution can be found in Usage of sample submission and Creating your own submission.

Non-compliance may result in premature termination of your solution with a Resource Limit Exceeded error.

Logs may be obtained only on a case-by-case basis. Requests can be made over at the discussion board, but the fulfilment of the request shall be at the discretion of the organisers.

Performance Metric Details

The performance metrics used are Area Under the Curve of the Receiver Operating Characteristic (AUROC) and accuracy. Both metrics will be displayed on the leaderboard.

1. The proba output is used to calculate AUROC and used to determine your ranking on the leaderboard.
2. The label output is used to calculate accuracy, and used as secondary metric in the event of a tie.

Docker Container Details

Max Docker Container Size

Your solution upon saving using docker save must not exceed the maximum file size of 25 GiB.

Max Docker Container Runtime

Your solution must not exceed 2.5 hours of runtime to process around 1700 images.

Submitted Docker Container Isolation

All submitted Docker containers are executed in a network isolated environment where there is no internet connectivity, nor access to any other external resources or data beyond what the container has.

As such, your solution must have all necessary modules, model weights, and other non-proprietary dependencies pre-packaged in your Docker container.

Non-compliance will result in your Docker container facing issues/error when in operation.

Example: Usage of sample submission

Pre-condition: Create the isolated Docker network

Before trying out the sample submission or creating your own submission, you will need to create a local Docker network to simulate the environment setup for the execution of solutions.

Run the following command to create your own isolated Docker network. If it is already created, as indicated by the output of docker network ls, you can skip this step.

ISOLATED_DOCKER_NETWORK_NAME=exec_env_jail_network
ISOLATED_DOCKER_NETWORK_DRIVER=bridge
ISOLATED_DOCKER_NETWORK_INTERNAL=true
ISOLATED_DOCKER_NETWORK_SUBNET=172.20.0.0/16

docker network create \
    --driver "$ISOLATED_DOCKER_NETWORK_DRIVER" \
    $( [ "$ISOLATED_DOCKER_NETWORK_INTERNAL" = "true" ] && echo "--internal" ) \
    --subnet "$ISOLATED_DOCKER_NETWORK_SUBNET" \
    "$ISOLATED_DOCKER_NETWORK_NAME"

Clone this repository and navigate to it

git clone https://github.com/AISG-Technology-Team/AISG-Online-Safety-Challenge-Submission-Guide.git

Change into the sample submission (sample_submission) directory

cd sample_submission

Build the sample Docker container

You can add a --no-cache option in the docker build command to force a clean rebuild.

docker build -t sample_container .

Please take note that the "." indicates the current working directory and should be added into the docker build command to provide the correct build context.

Test sample Docker container locally

Please ensure you are in the parent directory of sample_submission before executing the following command. The $(pwd) command in the --mount option yields the current working directory. The test is successful if no error messages are seen and a stdout.csv is created in the local_test/test_output directory.

Alter the options for --cpus, --gpus, --memory to suit the system you are using to test.

ISOLATED_DOCKER_NETWORK_NAME=exec_env_jail_network

cat local_test/test_stdin/stdin.csv | \
docker run --init \
        --attach "stdin" \
        --attach "stdout" \
        --attach "stderr" \
        --cpus 2 \
        --gpus "device=0" \
        --memory 4g \
        --memory-swap 0 \
        --ulimit nproc=1024 \
        --ulimit nofile=1024 \
        --network exec_env_jail_network \
        --read-only \
        --mount type=bind,source="$(pwd)"/local_test/test_images,target=/images,readonly \
        --mount type=tmpfs,destination=/tmp,tmpfs-size=5368709120,tmpfs-mode=1777 \
        --interactive \
        sample_container \
 1>local_test/test_output/stdout.csv \
 2>local_test/test_output/stderr.csv

Please note that the above docker run command would be equivalent to running the following command locally:

cat local_test/test_stdin/stdin.csv | \
    python3 sample_submission/main.py \
        1>local_test/test_output/stdout.csv \
        2>local_test/test_output/stderr.csv

(Optional) Adding own images

You can add a few more custom images (of the PNG format with file extension .png) into local_test/test_images. After doing that, you can regenerate the stdin.csv located in local_test/test_stdin using the following command.

cd utils

# Using the default location for the test_images and output_folder for storing stdin.csv
python3 gen_input.py --img_folder ../local_test/test_images --output_folder ../local_test/test_stdin

You can then re-run the Test sample Docker container locally

Compress your sample container to .tar.gz format using docker save

docker save sample_container:latest | gzip > sample_container.tar.gz

Upload container

The final step would be to submit the compressed Docker container file (sample_container.tar.gz in this example) on to the challenge platform, but since this is only the sample with no actual logic, we will not do so.

Please note that if you do submit this sample, it will still take up one count of your submission quota.

Example: Creating your own submission

The process of creating your own submission would be very similar to using the aforementioned sample submission.

Create a project directory and navigate into it

mkdir Online-Safety-Challenge && cd Online-Safety-Challenge

Create a main file

The main file has to be able to interact with standard streams such as stdin, stdout, and stderr.

In general, the main file should have the following characteristics:

1. Read the PNG images from the file paths obtained from stdin (one file path per input line);
2. Predict the probability that the image is a harmful meme, up to 4 decimal places (proba as referred to in the Submission Specification Guidelines);
3. Decide if the image is a harmful meme (1), or a benign meme (0) (label as referred to in the Submission Specification Guidelines);
4. Output a single line with to stdout for each line of stdin conforming to the Submission Specification Guidelines;
5. Use stderr to log any necessary exceptions/errors.

Note:

Please ensure that the prediction output to stdout is in the same order as the stdin because the order in which stdout is assessed strictly follows the order of the input from stdin.

You must use /tmp within your Docker container for any temporary files for processing. This is because the Docker container will be executed with the options:

• --read-only which sets the root file-system as read only.
• --tmpfs /tmp which sets a fixed /tmp directory for any app to write to.

You may refer to the main.py of the sample submission as an example of a main file.

Create a Dockerfile

You may use the sample Dockerfile provided for you. However, please install the relevant dependencies required for your detection model. Additionally, you may wish to change the ENTRYPOINT if you are using another main file or if you prefer to use a shell script:

ENTRYPOINT ["bash","/path/to/your/main.sh"]

If you are not familiar with how to build a Dockerfile, please refer to the official documentation for more information.

Build your Docker container using docker build

docker build -t your_container .

Please take note that the "." indicates the current working directory and should be added into the docker build command to provide the correct build context.

Test your Docker container locally

1. Create a test directory outside of your project directory

mkdir local_test && cd local_test

2. Create input and output directory within your test directory

mkdir test_images test_output

3. Add test images (PNG format with file extension .png) into test_images directory

4. Generate the input file (.csv).

You can use the following script utils/gen_input.py in this guide to generate the stdin.csv which will be used as the source for stdin.

# Using the default location for the test_images and output_folder for storing stdin.csv
python3 gen_input.py --image_folder ../local_test/test_images --output_folder ../local_test/test_stdin

The script will create a stdin.csv file within local_test/test_stdin which will be used as the source of input for stdin for your Docker container.

5. Test your container using docker run

Please ensure you are in the parent directory before executing the following command. The $(pwd) command in the --mount option yields the current working directory. The test is successful if no error messages are seen and a stdout.csv is created in the local_test/test_output directory.

Alter the options for --cpus, --gpus, --memory to suit the system you are using to test.

ISOLATED_DOCKER_NETWORK_NAME=exec_env_jail_network

cat local_test/test_stdin/stdin.csv | \
docker run --init \
        --attach "stdin" \
        --attach "stdout" \
        --attach "stderr" \
        --cpus 2 \
        --gpus "device=0" \
        --memory 4g \
        --memory-swap 0 \
        --ulimit nproc=1024 \
        --ulimit nofile=1024 \
        --network exec_env_jail_network \
        --read-only \
        --mount type=bind,source="$(pwd)"/local_test/test_images,target=/images,readonly \
        --mount type=tmpfs,destination=/tmp,tmpfs-size=5368709120,tmpfs-mode=1777 \
        --interactive \
        your_container \
 1>local_test/test_output/stdout.csv \
 2>local_test/test_output/stderr.csv

Compress your Docker container to .tar.gz format using docker save

docker save your_container:latest | gzip > your_container.tar.gz

Upload your container

Submit your your_container.tar.gz file onto the challenge platform. Please note that when you do this, it will take up one count of your submission quota.