spannotation

Description of software

The package introduces a novel approach to image annotation and segmentation, requiring users to select just three key points on an image. These points define a triangular area representing the region of interest, which the algorithm then uses to generate a binary mask. This mask distinctly categorizes the image into two classes: drivable and non-drivable regions.

Key applications of this package include:

1. Row Crop Management: In agricultural settings, such as cornfields, the package can segment row crops, identifying drivable paths for farm machinery and ensuring efficient navigation through the fields.
2. Off-Road Navigation: For off-road scenarios, like dirt tracks, the tool can demarcate navigable paths, assisting in the planning and navigation of off-road vehicles.
3. On-Road Navigation: In typical urban or rural roads, the package can be used to distinguish the actual road (drivable region) from its surroundings (non-drivable regions, including shoulders and adjacent land), aiding in basic navigation tasks.

This tool is particularly useful for researchers and practitioners in autonomous vehicle navigation, agricultural robotics, and geographic information systems (GIS), where accurate and efficient image segmentation is critical.Please read more about the project on the Github page.

Please check the package download statistics:

Time comparison with dense annotation

Spannotation saved average of 30 seconds on one annotation of an average crop row or dirt road.

spannotation: 6.03 seconds

Regular annotation: 40.39 seconds

Installation

To install SPAnnotation, simply use pip:

pip install spannotation

then to ensure the latest version is installed:

pip install --upgrade spannotation

or

If you want to install a specific version:

pip install spannotation==0.1.11

Usage on Python code Editor

Here's a quick example of how to use spannotation if you are running from a typical code editor like jupyter, VSCode, etc:

Step 1: Install spannotation

First, install the package using pip. Run the following command in a cell in your code editor: pip install spannotation

then to ensure the latest version is installed:

pip install --upgrade spannotation

or

If you want to install a specific version:

pip install spannotation==0.1.11

Step 2: Import the Package

In a new cell, import the MaskGenerator class from the package:

from Spannotation import MaskGenerator

Step 3: Initialize the Generator

Create an instance of the MaskGenerator: generator = MaskGenerator()

Step 4: Process Images

Now, you can use the generator to process an image or a folder of images.

To process a single image:

generator.process_image('path/to/your/image.jpg', 'path/to/save/mask')

To process all images in a folder:

generator.process_folder('path/to/your/images', 'path/to/save/masks')

Replace 'path/to/your/image.jpg', 'path/to/your/images', and 'path/to/save/masks' with the actual paths on your system.**

Usage on Command line / Terminal

Installation

pip install spannotation

then to ensure the latest version is installed:

pip install --upgrade spannotation

or

If you want to install a specific version:

pip install spannotation==0.1.9

After installing, you can use Spannotation to process images through the command line.

Process a Single Image

• To process a single image and generate a mask:
  1. Run the command: python3 -m Spannotation.cli
  2. Choose option 1 for a single image.
  3. Enter the full path to your image.
  4. Enter the full path where the mask should be saved.

Process Multiple Images in a Folder

• To process multiple images in a folder:
  1. Run the command: python3 -m Spannotation.cli
  2. Choose option 2 for a folder.
  3. Enter the full path to your folder containing images.
  4. Enter the full path where the masks should be saved.

Important Notice for GUI Window

When processing images, spannotation uses OpenCV to open a GUI window for point selection. Please note:

• After the first use, the OpenCV window might not automatically come to the foreground for subsequent image processing.
• If the image window does not appear in front, please manually click on the window from your taskbar or window manager.
• This behavior can vary based on your operating system and its window management settings.

Uninstalling spannotation

We hate to see you go but if you have to uninstalll for any reasons. Please use:

pip uninstall spannotation

This will remove the spannotation package from your Python environment.

Examples

Example 1: Row-Crop Segmentation

Example 2: Off-Road Navigation

Performing deep learning training using generated mask

Training performance of 1030 images/masks used to train a U-Net model:

Sample semantic segmentation from the model trained with the data:

Contributing

Contributions to spannotation are welcome! reach out to Folorunsho Samuel or help resolve the pending issues

License

spannotation is released under the MIT License.

Project details

Please check the gitup page : Github

Contact Info

Please reach out via:

LinkedIn

Email

Citation

If you use spannotation in your research or project, please consider citing it. Here is an example citation format you can use:

@misc{spannotation, author = {Samuel O. Folorunsho, William R. Norris}, title = {Spannotation: Enhancing Semantic Segmentation for Autonomous Navigation with Efficient Image Annotation}, year = {2024}, publisher = {arXiv}, journal = {cs.CV}, howpublished = {http://dx.doi.org/10.13140/RG.2.2.35355.85282 }