COCOSuite

COCOSuite is a comprehensive collection of tools designed to facilitate the management and manipulation of datasets in COCO format.

Tools

Tool	Description
coco_merge	Merge two COCO datasets into a single one
merge_multiple	Allows merging of multiple COCO files into a single dataset
coco_split	It consists of two functions, random_split performs a random division of the dataset into training and validation subsets, configurable in terms of data proportion, and property_split divides a COCO dataset into training and validation sets according to specific image properties
coco_filter	Filters a COCO dataset based on certain criteria
visualization	A series of visualization charts for analyzing and understanding data distributions, image sizes, annotation counts, and bounding box sizes within a dataset.

Installation

• Pip

  Install COCOSuite directly from PyPI:

  pip install cocosuite

• Source

  Clone the repository and install the dependencies:

  git clone https://github.com/jorgenusan/cocosuite.git
  cd cocosuite
  pip install -r requirements.txt

Usage

Important

In all scripts, if no path is specified in the <output_filename> argument and only a name is specified. The resulting file will be created in the input annotations file path

Basic example

python3 /cocosuite/scripts/manipulation/coco_merge.py <annotations_file_1> <annotations_file_2> <output_filename>

config_split

Applies to property_split and coco_filter

python3 /cocosuite/scripts/manipulation/property_split.py <annotations_file> <config_split>

For these scripts you have to pass a configuration file with the criteria for separating the data in the case of property_split and the criteria for filtering in the case of coco_filter.

Here is an example for each of these cases:

1. property_split

    "criteria": {
      "file_name": ["image1", "image2"],
      "height": [480]
    },
    "match_all": true

2. coco_filter

   "filter": {
     "file_name": ["image1", "image2"],
     "height": [480]
   },
   "match_all": true

Note

the match_all property, when set to true means that both properties have to match in order to filter or split a new file.
If set to false, it filters or splits for each property.