image-object-slicer 1.12.1

Slice images using annotation files.

image-object-slicer

Slice objects from images using annotation files. Convert an object detection dataset to an image classification one. To annotate the images to be used with this tool, we recommend openvinotoolkit/cvat.

This is a fork of gitlab.com/straighter/pascalvoc-to-image.

Installation

Install the latest stable version from PyPI with:

sudo pip3 install image-object-slicer

Or download the latest stable wheel file from the releases page and run:

sudo pip3 install ./image_object_slicer-*-py3-none-any.whl

Or install the latest development version from the git repository:

git clone https://www.github.com/natanjunges/image-object-slicer.git
cd image-object-slicer
sudo pip3 install ./

Usage

Different formats of annotation files are supported:

Annotation format	Command line option
MS COCO Object Detection	coco
CVAT for images	cvatimages
Datumaro	datumaro
KITTI	kitti
LabelMe	labelme
Open Images	openimages
Pascal VOC	pascalvoc
WIDER Face	widerface
YOLO	yolo

Using the script is pretty simple, since it only has three required parameters:

usage: image-object-slicer [-h] [-v] [-f {pascalvoc,coco,cvatimages,datumaro,kitti,labelme,openimages,widerface,yolo}] [-p PADDING] [-w WORKERS] annotations images save

Slice objects from images using annotation files

positional arguments:
  annotations           A path to the directory with the annotation files
  images                A path to the directory with the input images
  save                  A path to the directory to save the image slices to

options:
  -h, --help            show this help message and exit
  -v, --version         show program's version number and exit
  -f {pascalvoc,coco,cvatimages,datumaro,kitti,labelme,openimages,widerface,yolo}, --format {pascalvoc,coco,cvatimages,datumaro,kitti,labelme,openimages,widerface,yolo}
                        The format of the annotation files (default is pascalvoc)
  -p PADDING, --padding PADDING
                        The amount of padding (in pixels) to add to each image slice
  -w WORKERS, --workers WORKERS
                        The number of parallel workers to run (default is cpu count)

Building

To build the wheel file, you need deb:python3.10-venv and pip:build:

sudo apt install python3.10-venv
sudo pip3 install build

Build the wheel file with:

python3 -m build --wheel