stephenfang51/voc_to_coco

最简单的VOC转COCO， 一条指令完整转换

VOC_to_COCO

update 2019.10.5

最简单的VOC数据集转换为COCO数据集方式

修改路径后， 只要一条指令就能转换

任何报错请issue我

1. 请先确认你的VOC如下面的格式

     VOC
       |-- Annotations
       		|-- all xml files
       |-- JPEGImages
       		|-- all your samples
       |-- ImageSets
       		|-- 。。。。
   
   

2. 克隆本仓到你指定的地址

git clone https://github.com/Stephenfang51/VOC_to_COCO

3. cd 到VOC_to_COCO 编辑 voc_to_coco.py修改以下3点

   1. 设定验证集样本数, 如果设定为100， 则将从你的样本随机提取100个作为验证集
   2. 设定测试集样本数
   3. VOC Annotations 资料夹路径（最后面的"/"务必加上， 否则报错）

   例如

   val_files_num = 100
   test_files_num = 100
   voc_annotations = '././VOC/Annotations/'  #remember to modify the path

   修改到这边已经结束

4. 执行生成, 确认是python3版本以上解释器， 否则报错

   python3 voc_to_coco.py
   

5. COCO 数据集已经生成与VOC同一主目录下

---

The simplest way to covert VOC style dataset to COCO style dataset, only for object detection tesk for now.

Any problem feel free to issue me !

1. First need to comfirm that your VOC path looks like:

     VOC
       |-- Annotations
       		|-- all xml files
       |-- JPEGImages
       		|-- all your samples
       |-- ImageSets
       		|-- 。。。。
   

2. Clone this repo

   git clone https://github.com/Stephenfang51/VOC_to_COCO
   

3. command cd to VOC_to_COCO path, and set the following three parts

   val_files_num = 100
   test_files_num = 100
   voc_annotations = '././VOC/Annotations/'  #remember to modify the path

   ex.

   if you set val_files_num to 100, which means it wil randomly choice 100 samples from your dataset

4. do it ！ only for python3

   python3 voc_to_coco.py

5. Done ! your COCO dataset and your VOC dataset in the same path. So easy !

voc_to_coco's Issues

ValueError: Filename is supposed to be an integer.

[ Done ! Go check your file ]*
Traceback (most recent call last):
File "/home/litong/voc_to_coco/VOC_to_COCO/voc_to_coco.py", line 166, in get_filename_as_int
return int(filename)
ValueError: invalid literal for int() with base 10: ''

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "/home/litong/voc_to_coco/VOC_to_COCO/voc_to_coco.py", line 266, in
convert(xml_val_files, coco_json_annotations + 'val2017.json')
File "/home/litong/voc_to_coco/VOC_to_COCO/voc_to_coco.py", line 209, in convert
image_id = get_filename_as_int(filename)
File "/home/litong/voc_to_coco/VOC_to_COCO/voc_to_coco.py", line 168, in get_filename_as_int
raise ValueError("Filename %s is supposed to be an integer." % (filename))
ValueError: Filename is supposed to be an integer.

Process finished with exit code 1

请教坐标转换的问题，貌似有bug

左上角坐标转时候为什么要减掉1?而且右下角坐标没有减的话，算出来的宽度和高度会多一个像素啊

请问一下ImageSets里应该放什么文件

您好，我在使用您的代码是，我JPEGImages里有247个图片，ImageSets里不知道放什么所以空着在 然后用默认的100val 100test跑 最后结果train2017.jason是空的 然后python的最后一行如下
100 files copied to /Users/jayhuashi/Desktop/Sub22/xml/xml_val/
The folders are empty, please make sure there are enough 100 file to move

请问会是什么原因导致的呢 感谢

PermissionError: [Errno 13] Permission denied: '/VOC_to_COCO_COCO/'

请问这是什么问题

如果不设置数量，能不能自动选择把全部数据都转换完毕

谢谢你的代码，如果不设置数量，能不能自动选择把全部数据都转换完毕，比如官方原始的VOC2007和VOC2012

coco数据集图像不用分开？为什么xml分开了，但图片没有？

如何改动把图片也分开

concerning the file name error

Traceback (most recent call last):
File "voc_to_coco.py", line 166, in get_filename_as_int
return int(filename)
ValueError: invalid literal for int() with base 10: '000609af'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "voc_to_coco.py", line 266, in
convert(xml_val_files, coco_json_annotations + 'val2017.json')
File "voc_to_coco.py", line 209, in convert
image_id = get_filename_as_int(filename)
File "voc_to_coco.py", line 168, in get_filename_as_int
raise ValueError("Filename %s is supposed to be an integer." % (filename))
ValueError: Filename 000609af is supposed to be an integer.

Is this a file name error? Looking forward to your reply.

filename的id

我想转换通过labelImg标注的数据集为coco数据集，这个filename不是数字串，是字母等，在调用get_filename_as_int这个地方就会出错，这里是不是只是赋给每张图一个id，那我手动赋值可以吗

coco数据集图像不用分开？为什么xml分开了，但图片没有？

如何改动把图片也分开

[Errno 13] Permission Denied

Traceback (most recent call last):
File "voc_to_coco.py", line 65, in
mkdir(coco_path)
File "voc_to_coco.py", line 54, in mkdir
os.makedirs(path)
File "/usr/lib/python3.6/os.py", line 220, in makedirs
mkdir(name, mode)
PermissionError: [Errno 13] Permission denied: '/VOC2020_COCO/'

Firstly, I thought it may be the setting of Big Sur SIP. But after I change the setting on macOS or turning to other Operation System, the problem still exists.

显示Done但是依然报错

您好，我运行后您的代码后显示Done，但是同时也有报错的信息，请问是什么问题呢？

原始数据的目录（我用的是VOC2012）：

 VOC
    |-- Annotations
    		|-- 17125个xml
    |-- JPEGImages
    		|-- 17125张图片
    |-- ImageSets
    		|-- Action
    		|-- Layout
    		|-- Main
    		|-- Segmentation

 VOC_to_COCO
    |-- README.md
    |-- voc_to_coco.py

voc_to_coco.py中我只改了地址voc_annotations = '/home/vincent/Data/VOCdevkit/VOC/Annotations/'

生成的目录：

 VOC
    |-- Annotations
    		|-- 17125个xml
    |-- JPEGImages
    		|-- 17125张图片
    |-- ImageSets
    		|-- Action
    		|-- Layout
    		|-- Main
    		|-- Segmentation

 VOC_to_COCO
    |-- README.md
    |-- voc_to_coco.py

# 下面两个是新生成的

 VOC_COCO
    |-- annotations
    		|-- test2017.json
    		|-- val2017.json
    |-- images
    		|-- 17125张图片

 xml
    |-- xml_test
    		|-- 100个xml
    |-- xml_train
    		|-- 16925个xml
    |-- xml_val
    		|-- 100个xml

报错为：

/home/vincent/Data/VOCdevkit/VOC_COCO/ ----- folder created
/home/vincent/Data/VOCdevkit/VOC_COCO/images ----- folder created
/home/vincent/Data/VOCdevkit/VOC_COCO/annotations/ ----- folder created
/home/vincent/Data/VOCdevkit/xml/xml_val/ ----- folder created
/home/vincent/Data/VOCdevkit/xml/xml_test/ ----- folder created
/home/vincent/Data/VOCdevkit/xml/xml_train/ ----- folder created


 100 files copied to /home/vincent/Data/VOCdevkit/xml/xml_val/


***************************[ Done ! Go check your file ]****************************
Traceback (most recent call last):
  File "voc_to_coco.py", line 268, in <module>
    convert(xml_train_files, coco_json_annotations + 'train2017.json')
  File "voc_to_coco.py", line 231, in convert
    ymin = int(get_and_check(bndbox, "ymin", 1).text) - 1
ValueError: invalid literal for int() with base 10: '281.70000076293945'

我打开查看test2017.json和val2017.json，内容如下：
test2017.json内容：

images
    |-- 编号从0~99
type: "instances"
annotations
    |-- 编号从0~242
categories
    |-- 编号从0~19

val2017.json内容：

images
    |-- 编号从0~99
type: "instances"
annotations
    |-- 编号从0~199
categories
    |-- 编号从0~18

我修改了voc_to_coco.py中的val_files_num和test_files_num，尝试10或者1000，也是报同一个错误

我有以下的疑问：

1. 请问是不是应该还有一个关于train的json呢，因为报错的原因所以没有生成？
2. 请问有什么办法解决这个报错不？

谢谢！

coco数据集图像不用分开？为什么xml分开了，但图片没有？

如何改动把图片也分开

诶 怎么全网都找不到一个支持segment的