COCO-TOOLKIT

• Introduction

  • Preprocessing
  • Converter
  • Merge
  • Report

• Installation

• Usage

  • Import
    • Preprocess
    • Merge
    • Report
    • Converter
  • ClassPreProcess
    • reader
    • set_unique_image_id
    • set_unique_class_id
    • set_unique_annotation_id
    • check_id_unique
    • change_category_name_and_id
    • extract_data_by_class_name
    • separate_json_by_categories
    • filter_data_by_class_name
    • remove_segmentation
    • remove_distorted_bbox
    • box2segmentation
    • save_coco_file
    • compare_two_annotations
    • remove_duplicate_image_name
    • change_image_file_names
    • train_test_validation_split
    • unite_classes
    • image_split
  • FunctionMerge
    • merge_multiple_cocos
  • ClassAnalyzeCategories
    • Ground-truth
    • Paint-annotations
    • given_category_count
    • class_names
    • total_class_count
    • classes_have_annotations_tuple
    • plot_class_pie_chart
    • plot_category_destinations
    • images_aspect_ratio
    • bbox_aspect_ratio
    • class_have_ann_list
    • coco_viewer
  • FunctionConverter
    • voc2coco

• References

Introduction

Coco-toolkit is a tool for preparing and analyzing object detection data which is coco json format in Python. Tool countains merge, preprocessing, report and converter modules.

Preprocessing

This class obtain preprocess functions for preparing coco json dataset.

Converter

This module has converter function which is Pascal voc to coco json.

Merge

Merge module has multiple coco merge function. It merges all given coco json file and return all in one output folder.

Report

Report module has analyze dataset functions. These functions are; return information of data set, plots data set information as pie chart, and integrates data set with coco viewer.

Installation

In terminal code run;

pip install coco-toolkit

Usage

Import

---

Preprocess

from coco_toolkit.helper.preprocess import PreProcess

Merge

from coco_toolkit.helper.merge import merge_multiple_cocos

Report

from coco_toolkit.helper.report import AnalyzeCategories

Converter

from coco_toolkit.convertors.converter import voc2coco
from coco_toolkit.convertors.converter import coco2yolo

ClassPreProcess

---

reader

This function read given coco json file path . It returns coco json file as a dictionary.

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)

set_unique_image_id

This function set unique image id all images.

parameter first_id: First image id value

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
p = PreProcess(coco)
p.set_unique_image_id(first_id=1)
p.coco # Processed coco json file 

set_unique_class_id

This function set unique category id all categories.

parameter first_id: First image id value parameter b_grounds: If it's True add backgrounds to categories if there is no background category

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
p = PreProcess(coco)
p.set_unique_class_id(first_id=1-0, back_grounds=True)
p.coco # Processed coco json file 

set_unique_annotation_id

This function set unique annotation id all annotations.

parameter first_id: First image id value

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
p = PreProcess(coco)
p.set_unique_annotation_id(first_id=0-1)
p.coco # Processed coco json file 

check_id_unique

This function check annotations, image and category id. If all id are unique return True, if id not unique return False

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
p = PreProcess(coco)
result = p.check_id_unique() # Return False or True

change_category_name_and_id

This function

param path: json path param object_id: old category id param new_id: new category id param new_name: new category name

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
p = PreProcess(coco)
p.change_category_name_and_id(object_id=1, new_id=3, new_name="cat")

extract_data_by_class_name

This function extract coco json file according to given list of categories. Saves new coco json data(annotations and images) to given out_path.

parameter categories: List of chosen categories names parameter image_path: Image path of dataset parameter out_path: Output directory

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
img_path = "coco_dataset/images"
out_path = os.getcwd()
p = PreProcess(coco)
export_list = ["crosswalk"]
extract_list = ["crosswalk"] # Category to be extracted from coco data
p.extract_data_by_class_name(
            categories=export_list, image_path=img_path,
            out_path=out_path
        )
p.coco # Processed coco json file 

separate_json_by_categories

This function separate json by categories

For example:

path = "/coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
p = PreProcess(coco)
p.separate_json_by_categories()

filter_data_by_class_name

This function remove categories by given list of category names. Saves new coco json data(annotations and images) to given out_path.

parameter categories: List of chosen categories names parameter image_path: Image path of dataset parameter out_path: Output directory

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
img_path = "coco_dataset/images"
out_path = os.getcwd()
p = PreProcess(coco)
filter_list = ["crosswalk"]
p.filter_data_by_class_name(
            categories=filter_list, image_path=img_path,
            out_path=out_path
        )
p.coco # Processed coco json file 

remove_segmentation

This function remove segmentations from annotations. Return updated coco json file. If parameter inplace True save updated coco json file to given path.

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
p = PreProcess(coco)
p.remove_segmentation()
p.coco # Processed coco json file 

remove_distorted_bbox

This function remove distorted bbox from annotations.

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
p = PreProcess(coco)
p.remove_distorted_bbox()
p.coco # Processed coco json file 

box2segmentation

This function create segmentation list from bbox to annotations if there is no segmentation list .

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
p = PreProcess(coco)
p.box2segmentation()
p.coco # Processed coco json file 

save_coco_file

This function save coco json file to given directory + file name.

parameter directory: The directory of coco json file to be saved parameter file_name: The file name of coco json file to be saved(without extencion)

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
p = PreProcess(coco)
dir_ = os.getcwd()
f_name = "test_save"
p.save_coco_file(directory=dir_, file_name=f_name)

remove_duplicate_image_name

This function if there is a duplicate image name in coco json file remove duplicate name.

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
p = PreProcess(coco)
p.remove_duplicate_image_name()
p.coco # Processed coco json file 

compare_two_annotations

This function compare annotations two json

For example:

path1 = "coco_dataset/annotations/coco.json"
path2= "coco_dataset_1/annotations/coco.json"
coco = PreProcess.reader(path)
p = PreProcess(coco)
p.compare_two_annotations(path1=path1,path2=path2)

change_image_file_names

This function change images' file name and copy them to a new folder. If inplace True save coco json file to new folder.

parameter image_path: Path of folder that contains dataset images parameter inplace: If it's True create new coco json file to given directory

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
p = PreProcess(coco)
img_path = "coco_dataset/images"
p.change_image_file_names(image_path=img_path, inplace=True)
p.coco # Processed coco json file 

train_test_validation_split

This function split data set to train, test and validation according to given split percent. Return train,test and validation coco json file.

parameter image_path: Path of folder that contains dataset images parameter test_percent: Test split percent parameter validation_percent: Validation split percent parameter out_path: Output path

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
p = PreProcess(coco)
img_path = "coco_dataset/images"
output_path = "/home/documents"

train,test,validation = p.train_test_validation_split(image_path=img_path, test_percent=20, validation_percent=15, out_path=output_path)

unite_classes

This function unite given classes in a class.

parameter class_names: List of class names parameter new_class_name: Name of class name to be created

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
p = PreProcess(coco)
class_name_list = ["stop", "trafficlight", "crosswalk"]
new_class_name = "Road Sign"
p.unite_classes(class_names=class_name_list, new_class_name=new_class_name)
p.coco # Processed coco json file 

image_split

This functon split image dataand create train test validation image folders.

parameter image_path: Path of folder that obtain images parameter test_percent: Image split test percent parameter val_percent: Image split val percent

For example:

img_path = "coco_dataset/images"
PreProcess.image_split(image_path=img_path, test_percent=15, val_percent=25)

FunctionMerge

---

merge_multiple_cocos

This function merge all given coco datasets and save given directory.

parameter args: It contains lists in index 0 json path and index 1 images path. For example [json_path_1, image path_1], [json_path_2, image path_2] parameter merge_path: Path of output folder directory parameter first_id: Value of first id parameter visualizer: if it's True visualize categories with pie chart

For example:

path_1 = "coco_dataset_1/annotations/coco.json"
img_1 = "coco_dataset_1/images"
path_2 = "coco_dataset/annotations/coco.json"
img_2 = "coco_dataset/images"
merge_path = "test_merge"
args = [path_1, img_1], [path_2, img_2]
merge = merge_multiple_cocos(*args, merge_path=merge_path, first_id=100000, visualizer=True)

ClassAnalyzeCategories

---

given_category_count

This function return count of given category annotations.

parameter category: Category name

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
count = AnalyzeCategories(coco).given_category_count("crosswalk")

class_names

This function return list of class names .

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
class_name_list = AnalyzeCategories(coco).class_names()

Ground-truth

For example:

path = "coco_dataset/annotations/coco.json"
pyodi ground-truth (path)

Paint-annotations

For example:

jsonpath = "coco_dataset/annotations/coco.json"
images= "coco_dataset/images/"
painted_images= "output"
pyodi paint-annotations (jsonpath) (images) (painted_images)

total_class_count

This function return class count.

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
count = AnalyzeCategories(coco).total_class_count()

plot_category_destinations

This function draw category destinations For example:

path = "tests/coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
tuple = AnalyzeCategories(coco).plot_category_destinations()

classes_have_annotations_tuple

This function return class names as tuple that has annotation.

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
tuple = AnalyzeCategories(coco).classes_have_annotations_tuple()

plot_class_pie_chart

This function save class information pie chart as png and if parameter visualize True visualize pie chart

parameter visualize: If it's True visualize pie chart.

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
AnalyzeCategories(coco).plot_class_pie_chart(visualize=True)

images_aspect_ratio

This function return image dataset aspect ratio as dictionary.

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
info_dictionary = AnalyzeCategories(coco).images_aspect_ratio()

bbox_aspect_ratio

This function return dictionary of bbox aspect ratio information. It returns three aspect ratio which is has max count.

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
info_dictionary = AnalyzeCategories(coco).bbox_aspect_ratio()

class_have_ann_list

This function return list of class names that has annotation.

For example:

path = "coco_dataset/annotations/coco.json"
coco = PreProcess.reader(path)
list_class_names = AnalyzeCategories(coco).class_have_ann_list()

coco_viewer

This function open coco viewer .

For example:

path = "coco_dataset/annotations/coco.json"
img_path = "coco_dataset/images"
AnalyzeCategories.coco_viewer(image_path=img_path, json_path=path)

FunctionConverter

---

voc2coco

This fuction convert poscal voc format to coco json format. And return coco json file as dictionary and saves coco data set in given output path.

parameter data_xml_folder_path: directory of folder that obtain datas in format xml parameter output_path: directory of folder that created coco json parameter image_path: Data set's images path

path_xml = "coco_dataset/annotations_xml"
img_path = "xml_dataset/images"
output_path = "/home/documents"
voc2coco(data_xml_folder_path=path_xml, output_path=output_path, image_path=img_path)

References

https://github.com/trsvchn/coco-viewer

https://github.com/yukkyo/voc2coco

`

Check before PR

black . --config pyproject.toml
isort .
pre-commit run --all-files