Visit simple-tools-for-machine-learning for a newer tool, which written in python3.8,skimage,numpy and without opencv-python. This repo will not be upgraded any more.

convertmask

Issues and advices wanted.

Introduction

A small tool for image augmentation, including mask files to json/xml files , image augmentation(flip,rotation,noise,...) and so on

HOW TO USE.

Installation.

Try:

pip install -r requestments.txt

IF ERROR, try:

pip install -r requirements.txt --ignore-installed

Use.

Under this version, these tools are provided.

1.augmentation

2.img2xml

3.json2mask

4.json2xml

5.longImgSplit

6.xml2json

7.xml2mask

8.xml2yolo

9.yolo2xml

10.mask2json

CHANGE LOGS

2021.10.31

1.bump to 0.6.0

2. UI update

2021.1.14

1.update spliting yolo-like dataset into train and val datasets AUTOMATICLY.See testYoloLike.py. The test dataset can be found in ./test/ , 68 txts totally.

2020.11.25

1. negative sample AUTOMATICLY randomly generate. I am not sure if it is helpful to my face detection cnn.

2. rewrite mosaic.py(without resize the origin images). See here.

2020.11.16

1.support converting widerface annotations to xmls. see here

2020.11.10

1. object-oriented rewrite augumentation optional module. See here. Test script can be found here

2020.10.25

1. bump to 0.5.3

2. code structure change

3. mosiac(yolov4) augumentation supported(auto labeled for labelImg,for labelme will be updated as fast as i can). Script can be found here

2020.10.24

1.image crop supported.(single and multiple crops,rectangle and polygon support. See here)

2.image resize supported (auto labeled). See here and the test script is here

2020.10.23

1.image distortion supported.

see here or test-script for details.

2020.10.13

1.image augumentation support convert yolo txts to xmls(pascal). See here

2.speed up by using multiprocess

2020.10.12

1.image augumentation support generating several annotation/images with single image/annotation(json,xml). See here

2020.9.24

1.inspired by LabelImgTool, convert xmls to jsons is useful. Also ,i forked this repo and add some pyqt5/py3 support,see here.

examples:

script here

2020.8.24

1.support convert xml files to yolo files. see here

examples:

2020.8.19

1. image translation supported.

combination of every augmentation method.

2. besides, a simple way convert json file(labelme) to xml file(labelImg) is provided. see here

2020.8.17

1. bug fix.

2. support image augmentation methods: noise,flip,rotation. try test_imgAug.py !

here are some examples:

flip

noise

rotation

2020.8.14

1. add image augmentation (image flip) test. see test_imgAug.py !

2020.7.14

1.bugfix , test multi objects to xml files, pretty xmls

eg:

2020.7.13

1. convert multi objects to xml files supported (untested)

2020.7.10

1. a lot of things to do ,such as many warnings related to labelme.

2020.6.12

1.support multiple objects mask to json

try test.py !

1.1 multiple objects in different classes

manually_labeled image

auto_labeled image

1.2 multiple objects in same classes

manually_labeled image

auto_labeled image

what to do next

1. support multiple files image augmentation (2020.8.21)

2. support image augmentation without a label/json file (2020.8.21)

3. support image augmentation with a labeled file (just support json file right now) (2020.9)

4. image augmentation supports custom parameters (auto augmented right now)

5. do something more interesting

6. re-write main script (2020.10)

7. ~~ solve PyYmal installation error(currently write a script by myself. maybe failure in the end.)~~

8. image augmentation zoooom (2020.10.14)

9. yolo txts split train/val dataset automaticly.

OTHERS

1. issue:json2mask has no output

if you test the json2mask.py script, you should change the path first and make sure the file is valid(maybe i have deleted :) )

Also, this script is just a reverse of mask2json, for a more COOOOOL method, see here , try convert.processor

SHORTCOMING

1.objects connected to each other is not supported yet.

this may happen if you labelling multiple-object-images with only 2 labels .Or some objects are of the same type and are connected to each other(eg. a bunch of grapes,it is hard to split one to the other).

2.Image Binarization issue