XML to JSON converter and polygon shape builder for MASK-RCNN

Annotation builder to use segmentation in Mask_RCNN, even if your annotations are rectangular instead of polygon. A false masks does not help to be more precise, as it is not providing new information beyond the bounding boxes. Generates the polygon shape between the four points of the rectangle in order to achieve the polygon shape.

Requirements

Python >=3.6

Usage

/dataset
- /train
- /val
main.py

$ python main.py

How should your XML be?

You can check this xml and see the result JSON.

<annotation>
	<folder>keuring</folder>
	<filename>IMG_20180413_091455.jpg</filename>
	<path>D:\Seafile\Archive\Projects\125 IWAUC\IMG_20180413_091455.jpg</path>
	<source>
		<database>Unknown</database>
	</source>
	<size>
		<width>3968</width>
		<height>2976</height>
		<depth>3</depth>
	</size>
	<segmented>0</segmented>
	<object>
		<name>SD</name>
		<pose>Unspecified</pose>
		<truncated>0</truncated>
		<difficult>0</difficult>
		<bndbox>
			<xmin>2241</xmin>
			<ymin>997</ymin>
			<xmax>2713</xmax>
			<ymax>1325</ymax>
		</bndbox>
	</object>
</annotation>

to achieve this

{
    "IMG_20180413_091455.jpg": {
        "filename": "IMG_20180413_091455.jpg",
        "regions": {
            "0": {
                "name": "polygon",
                "region_attributes": {
                    "name": "SD"
                },
                "shape_attributes": {
                    "all_points_x": [
                        2241, # xmin
                        2477, # X
                        2713, # xmax
                        2713, # xmax
                        2713, # xmax
                        2477, # X
                        2241, # xmin
                        2241, # xmin
                        2241  # xmin
                    ],
                    "all_points_y": [
                        997,  #ymin
                        997,  #ymin
                        997,  #ymin
                        1161, #Y
                        1325, #ymax
                        1325, #ymax
                        1325, #ymax
                        1161, #Y
                        997   #ymin
                    ]
                }
            }
        },
        "size": 1231831
    }
}

keyworks

xml, json, converterter, xml to json, convertidor de xml a json, mask rcnn

X = xmin + ((xmax-xmin)/2)
Y = ymin + ((ymax-ymin)/2)

The way to create a false mask for MaskRCNN with annotations made with LabelImg.

Author

Adonis González Godoy (Email - Github)

Assertion error-- Path file issue

while trying to convert my xml file to json I am getting the following error:

Traceback (most recent call last):
File "main.py", line 82, in
remove_file(file_train)
File "/home/hiwi/Auto-Annotate/xml2VIA/util.py", line 67, in remove_file
assert isfile(file_name), "-- check your path file"
AssertionError: -- check your path file

I have also editted the main.py file path accordingly like this:

`"""
converterXMLtoJSON

Detect annotations in xml files (bounding boxes) and
converts to json (polygon shape).

@author Adonis Gonzalez

"""
from os.path import dirname, realpath
import xml.etree.cElementTree as et
from util import *
import json

Paths

ROOT_DIR = '/path/to/xmlfile/train.xml'
data_dir = '/path/to/images'
train_dir ='/path/to/train'
#val_dir = join(data_dir, "val")

Files to create, just path definition

filename_json = "VIA4Annotate.json"
file_train = join(train_dir, filename_json)
#file_val = join(val_dir, filename_json)

def convert_xml_to_json(path: str, image_list: list):
"""
Convert from xml to json. For each img is necessary
a xml file annotation. This function save json file.

:param path: A str like /to/path/file
:param image_list: A list of images in the same path
"""
all_json = {}

for img in image_list:
    name_xml = img.split('.jpg')[0] + '.xml'
    images = ({"filename": img})
    root = et.ElementTree(file=join('/path/to/xml file', train.xml)).getroot()
    obj_counter, regi = {}, {}
    number = 0
    for child_of_root in root:
        if child_of_root.tag == 'object':
            for child_of_object in child_of_root:
                if child_of_object.tag == 'name':
                    obj_id = child_of_object.text.split(' ')[0]  # cause some <name>SD 1<name>, just use SD
                    obj_counter[obj_id] = number
                if child_of_object.tag == 'bndbox':
                    for child_of_root in child_of_object:
                        if child_of_root.tag == 'xmin':
                            x_min = int(child_of_root.text)
                        if child_of_root.tag == 'xmax':
                            x_max = int(child_of_root.text)
                        if child_of_root.tag == 'ymin':
                            y_min = int(child_of_root.text)
                        if child_of_root.tag == 'ymax':
                            y_max = int(child_of_root.text)

            x_value, y_value = calculate_xy(x_max, x_min, y_max, y_min)
            coord = get_points(x_max, x_min, y_max, y_min, x_value, y_value)

            regions = ({"region_attributes": {"name": obj_id}})
            regions.update({"shape_attributes": coord})
            regions.update({"name": "polygon"})
            regi[number] = regions.copy()
            regions = {"regions": regi}
            images.update(regions)
            images.update({"size": os.path.getsize(join(path, img))})
            all_json[img] = images.copy()
            number += 1

out_file = open(join(path, filename_json), "a")
json.dump(all_json, out_file)
print("File dataset.json was save in: ", path)

if name == "main":
# I convert for both train and val dataset annotation.
# If you just have a folder, just comment step for val.
# If not file dataset.json, just comment next two lines.
remove_file(file_train)
#remove_file(file_val)

# Grab images and save a log in both train and val
images_train = save_images_log(train_dir)
#images_val = save_images_log(val_dir)

# Convert from xml to json in both train and val
convert_xml_to_json(train_dir, images_train)
#convert_xml_to_json(val_dir, images_val)

# json1 = json.dumps(read_json(train_dir, "dataset.json"), sort_keys=True)
# json2 = json.dumps(read_json(train_dir, "dataset_good.json"), sort_keys=True)
# if json1 == json2:
#     print("Equals!")`

Can you please tell me what to do? Thanks.

adions025 / xmltojson_mask_rcnn Goto Github PK

xmltojson_mask_rcnn's Introduction

XML to JSON converter and polygon shape builder for MASK-RCNN

Requirements

Usage

How should your XML be?

to achieve this

keyworks

Author

xmltojson_mask_rcnn's People

Contributors

Stargazers

Watchers

Forkers

xmltojson_mask_rcnn's Issues

Paths

Files to create, just path definition

Recommend Projects

Recommend Topics

Recommend Org