Problem Statement

The Problem Statement, given by Adobe as part of Inter IIT Tech Meet 13.0 addresses challenges in the domains of Image Classification, Artefact Detection and Explainable AI. We aim to accomplish the following tasks:

• Detect AI-generated images.
• Identify artefacts in AI-generated images and generate appropriate explanations.

Getting Started

• The required python scripts for the solution will be executed using the Visual Studio (VS) Code IDE (recommended) or in the terminal window itself. Running these scripts requires the following steps to be completed:

  • Installation of a python interpreter, which supports a Python version of 3.10 or earlier.

  • Installation of the VS Code Python extension.

  • Creation of a workspace folder, via the execution of the mkdir command in the terminal.

      mkdir adobe_team_97

• The user changes directory to the folder by running the cd command, followed by the creation of a virtual environment (venv) named venv.

      cd adobe_team_97
      python3 -m venv venv

• Enable Long Path Support on Windows:

  • Press Win + R, type regedit, and press Enter to open the Registry Editor.
  • Navigate to the following path:

      HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\FileSystem

  • Find the key named LongPathsEnabled.
  • Double-click it and set its value to 1.
  • Click OK and close the Registry Editor.

• Activate the virtual environment by running the appropriate command based on your operating system.

  • Windows:

    venv\Scripts\activate.ps1

  • macOS/Linux/WSL/Git Bash:

    source venv/bin/activate

• Select the Virtual Environment:

  • In VS Code, click on the Python interpreter displayed in the bottom-left corner of the window (or use Ctrl + Shift + P and select "Python: Select Interpreter").
  • Choose the interpreter from the virtual environment folder (e.g., venv).

Installation

• The submitted folder needs to be installed and a copy of the folder must be saved in the adobe_team_97 directory.

• Change user directory to the downloaded sub-folder using the cd command.

    cd ./<sub_folder>

• Ensure that pip is upgraded to the latest version

    python.exe -m pip install --upgrade pip

• The required dependenices can be installed by running the requirements.txt file using the pip command.

    python3 -m pip install --no-cache-dir -r .\requirements.txt

• Create a directory named gradcam_images using the mkdir command

    mkdir gradcam_images

Execution

• Run the Inference Notebook from the VS Code terminal by executing the python3 command.

    python3 adobe_team97_inference_notebook.py

• Run the app.py file via the python3 command to run flask.

    python3 app.py

Outputs and Results

• The output printed in the terminal constitutes the results printed in the following format.

    Grad-CAM visualization saved at grad_cam_output_32x32.jpg
    Processed <image_path>:{image_class}

• The parameter image_class is either real or fake based on the authenticity of the given image.

• The complete results of the predictions for task 1 are saved in a .json file named task1_predictions.json.

• The complete results of the predictions for task 2 are saved in a .json file named task2_explanations.json.