GitHub - Giskard-AI/giskard-vision: 📸 Open-Source Evaluation & Testing for Computer Vision AI systems

The testing framework dedicated to Computer Vision models.

Detect risks of biases, performance issues and errors in your computer vision models.

Documentation • Blog • Website • Discord Community • Advisors

Giskard Vision is a comprehensive Python package designed to simplify and streamline a variety of computer vision tasks. Whether you're working on facial landmark detection, image classification, or object detection, Giskard Vision provides the tools you need to evaluate your models with ease.

Getting Started

To get the most out of Giskard Vision, we recommend starting with these essential guides in our documentation:

• Quickstart Guide
• Scan Overview

Supported Computer Vision Tasks

• Facial Landmark Detection (Readme)
• Image Classification (Readme)
• Object Detection (Readme)

Installation

To install Giskard Vision, simply use pip:

pip install giskard-vision

If you want to contribute to the development or explore the latest features, you can install the repository in development mode:

git clone https://github.com/Giskard-AI/giskard-vision.git
cd giskard-vision
pdm install -G :all
source .venv/bin/activate

Scan

Giskard Vision includes powerful scanning capabilities to evaluate your models. To run a scan, first ensure that you have the giskard library installed:

Then, you can perform a scan using the following code:

from giskard_vision import scan

dataloader = ...
model = ...

results = scan(model, dataloader)

Explore the examples provided to see how to implement scans in different contexts:

• Facial Landmark Detection Notebook
• Image Classification Notebook
• Object Detection Notebook

Examples

The examples directory contains Jupyter notebook tutorials that demonstrate how to use Giskard Vision for various tasks. To explore these tutorials:

1. Install the repository in development mode.
2. Navigate to the examples directory and open the notebook of interest.

FAQ

→ I am getting attributeerror: module 'cv2.face' has no attribute 'createlbphfacerecognizer' when running some examples in dev mode

This issue usually occurs due to the installation order of the opencv-contrib-python module. To resolve it, follow these steps:

pip uninstall opencv-contrib-python
pip install opencv-contrib-python

→ For Linux users with CUDA support

It is recommended that you install the following CUDA-compatible versions of Torch by running the command below:

pdm run pip install -U torch==2.1.0+cu121 torchvision==0.16.0+cu121 --index-url https://download.pytorch.org/whl/cu121