Data Format Converter

Deep learning dataset format converter. Convert your dataset in coco, voc, yolo format to other formats.

Usage

python convert.py --input_format voc/yolo/coco --input_dir data_yolo --output_format voc/yolo/coco --output_dir data_coco --subset train --class_file classes.txt --copy_img false

Args:
--input_format  input dataset format, one of (voc, yolo, coco)
--input_dir     input dataset path
--output format output dataset format, one of (voc, yolo, coco)
--output_dir    output dataset path
--subset        subset to convert
--class_file    class list file
--copy_img      whether to copy images or create soft link of images

YOLO

Data structure of yolo format:

--classes.txt
--train
   |--001.jpg
   |--001.txt
   |--002.jpg
   |--002.txt
--val
   |--101.jpg
   |--101.txt
   |--102.jpg
   |--102.txt

• classes.txt contains names of classes. One class name per line,

person
cat
dog

• images folder contain images
• labels folder contain labels. Label file and image file should have the same filename except extension. Labels are storing in the following format:

class_idx center_x center_y width height

One object per line. center_x, center_y, width, height are values in [0, 1]. Absolute values divided by the image width and height.

Pascal Voc

Data structure of voc format:

--ImageSets
  |--Main
     |--train.txt
     |--val.txt
--JPEGImages
   |--001.jpg
   |--002.jpg
   |--101.jpg
   |--102.jpg
--Annotations
   |--001.xml
   |--002.xml
   |--101.xml
   |--102.xml

• train.txt and val.txt contain the filenames of train and val split. One filename per line, without extension, like this:

001
002

• JPEGImages contains all the image files.
• Annotations contains all the annotation files. XML file sample:

<annotation>
    <filename>001.jpg</filename>
    <size>
        <width>1920</width>
        <height>1080</height>
        <depth>3</depth>
    </size>
    <object>
        <name>person</name>
        <bndbox>
            <xmin>100</xmin>
            <ymin>100</ymin>
            <xmax>200</xmax>
            <ymax>500</ymax>
        </bndbox>
    </object>
</annotation>

COCO

Data structure of coco format:

--images
  |--001.jpg
  |--002.jpg
--annotations
  |--train.json
  |--val.json

• images folder contains all the image files.
• annotations folder contains annotation files. Sample json file:

{
    "images": [{"file_name": "001.jpg", "id": 1, "width": 1920, "height": 1080}], 
    "annotations": [{"image_id": 1, "id": 1, "category_id": 1, "bbox": [100, 100, 100, 400]}], 
    "categories": [{"name": "person", "id": 1}]
}

bbox values are [left, top, width, height].