/convert2Yolo

This project purpose is convert voc annotation xml file to yolo-darknet training file format

Primary LanguagePython

Convert2Yolo

Object Detection annotation Convert to Yolo Darknet Format

Support DataSet :

  1. COCO
  2. VOC
  3. UDACITY Object Detection
  4. KITTI 2D Object Detection

Pre-Requiredment

pip3 install -r requirements.txt

Required Parameters

each dataset requried some parameters

see example.py

  1. --datasets

    • like a COCO / VOC / UDACITY / KITTI

      --datasets COCO
  2. --img_path

    • it directory path. not file path

      --img_path ./example/kitti/images/
  3. --label

    • it directory path. not file path

      (some datasets give label *.json or *.csv . this case use file path)

      --label ./example/kitti/labels/

      OR

      --label ./example/kitti/labels/label.json
      
      or
      
      --label ./example/kitti/labels/label.csv
  4. --convert_output_path

    • it directory path. not file path

      --convert_output_path ./
  5. --img_type

    • like a *.png, *.jpg

      --img_type ".jpg"
  6. --manifest_path

  7. --cla_list_file(*.names)

*.names file example

aeroplane
bicycle
bird
boat
bottle
bus
car
cat
chair
cow
diningtable
dog
horse
motorbike
person
pottedplant
sheep
sofa
train
tvmonitor

Example

1. example command

python3 example.py --datasets [COCO/VOC/KITTI/UDACITY] --img_path <image_path> --label <label path or annotation file> --convert_output_path <output path> --img_type [".jpg" / ".png"] --manifest_path <output manipast file path> --cls_list_file <*.names file path>

>>
ex) python3 example.py --datasets KITTI --img_path ./example/kitti/images/ --label ./example/kitti/labels/ --convert_output_path ./ --img_type ".jpg" --manifest_path ./ --cls_list_file names.txt

2. VOC datasets

description of dataset directory

suppose that VOC dataset location are ~/VOC and VOC folder contains VOCdevkit folder

here are structure for VOCdevkit

VOCdevkit

$ tree -L 2
.
└── VOC2012
    ├── Annotations
    ├── ImageSets
    ├── JPEGImages
    ├── SegmentationClass
    └── SegmentationObject

we use only Annotations and JPEGImages folder

  • Annotations : Object Detection label folder
  • JPEGImages : Image data

Annotations

$ tree -L 1
.
├── 2007_000027.xml
├── 2007_000032.xml
├── 2007_000033.xml
...
├── 2012_004319.xml
├── 2012_004326.xml
├── 2012_004328.xml
├── 2012_004329.xml
├── 2012_004330.xml
└── 2012_004331.xml

JPEGImages

.
├── 2007_000027.jpg
├── 2007_000032.jpg
├── 2007_000033.jpg
...
├── 2012_004328.jpg
├── 2012_004329.jpg
├── 2012_004330.jpg
└── 2012_004331.jpg

make *.names file

now make *.names file in ~/VOC/

refer darknet voc.names file

aeroplane
bicycle
bird
boat
bottle
bus
car
cat
chair
cow
diningtable
dog
horse
motorbike
person
pottedplant
sheep
sofa
train
tvmonitor

VOC datasets convert to YOLO format

now execute example code.

this example assign directory for saving YOLO label ~/YOLO/ and assign manifest_path is ./

make YOLO folder

$ mkdir ~/YOLO

VOC convert to YOLO

python3 example.py --datasets VOC --img_path ~/VOCdevkit/VOC2012/JPEGImages/ --label ~/VOCdevkit/VOC2012/Annotations/ --convert_output_path ~/YOLO/ --img_type ".jpg" --manifest_path ./ --cls_list_file ~/VOC/voc.names

>>
VOC Parsing:   |████████████████████████████████████████| 100.0% (17125/17125) Complete
YOLO Generating:|████████████████████████████████████████| 100.0% (17125/17125)Complete
YOLO Saving:   |████████████████████████████████████████| 100.0% (17125/17125) Complete

Result

now check result files (~/YOLO/, ./manifest.txt)

~/YOLO/

$ tree -L 1
>>
├── 2012_004326.txt
├── 2012_004328.txt
├── 2012_004329.txt
├── 2012_004330.txt
└── 2012_004331.txt
...
├── 2012_004326.txt
├── 2012_004328.txt
├── 2012_004329.txt
├── 2012_004330.txt
└── 2012_004331.txt

2012_004331.txt

$ cat 2012_004331.txt

>>
14 0.31 0.34 0.212 0.547

./manifest.txt

$ cat ./manifest.txt

>>
~/VOC/VOCdevkit/VOC2012/JPEGImages/2010_000420.jpg
~/VOC/VOCdevkit/VOC2012/JPEGImages/2010_003674.jpg
~/VOC/VOCdevkit/VOC2012/JPEGImages/2012_002128.jpg
...
~/VOC/VOCdevkit/VOC2012/JPEGImages/2009_000104.jpg
~/VOC/VOCdevkit/VOC2012/JPEGImages/2012_000212.jpg

3. COCO datasets

description of dataset directory

suppose that COCO dataset location are ~/COCO and COCO folder contains annotations, val2017 folder

here are each structure for annotations and val2017

annotations

$ cd ~/COCO/annotations/
$ tree -L 1
.
└── instances_val2017.json

val2017

.
├── 000000000139.jpg
├── 000000000285.jpg
├── 000000000632.jpg
├── 000000000724.jpg
...
├── 000000581357.jpg
├── 000000581482.jpg
├── 000000581615.jpg
└── 000000581781.jpg

make *.names file

now make *.names file in ~/COCO/

refer darknet coco.names file

person
bicycle
car
motorbike
aeroplane
bus
train
truck
boat
traffic light
fire hydrant
stop sign
parking meter
bench
bird
cat
dog
horse
sheep
cow
elephant
bear
zebra
giraffe
backpack
umbrella
handbag
tie
suitcase
frisbee
skis
snowboard
sports ball
kite
baseball bat
baseball glove
skateboard
surfboard
tennis racket
bottle
wine glass
cup
fork
knife
spoon
bowl
banana
apple
sandwich
orange
broccoli
carrot
hot dog
pizza
donut
cake
chair
sofa
pottedplant
bed
diningtable
toilet
tvmonitor
laptop
mouse
remote
keyboard
cell phone
microwave
oven
toaster
sink
refrigerator
book
clock
vase
scissors
teddy bear
hair drier
toothbrush
motorcycle
potted plant
dining table
tv
couch
airplane

COCO datasets convert to YOLO format

now execute example code.

this example assign directory for saving YOLO label ~/YOLO/ and assign manifest_path is ./

make YOLO folder

$ mkdir ~/YOLO

COCO convert to YOLO

python3 example.py --datasets COCO --img_path ~/COCO/val2017/ --label ~/COCO/annotations/instances_val2017.json --convert_output_path ~/YOLO/ --img_type ".jpg" --manifest_path ./ --cls_list_file ~/COCO/coco.names

>>
COCO Parsing:  |████████████████████████████████████████| 100.0% (36781/36781) Complete
YOLO Generating:|████████████████████████████████████████| 100.0% (4952/4952)  Complete
YOLO Saving:   |████████████████████████████████████████| 100.0% (4952/4952)  Complete

Result

now check result files (~/YOLO/, ./manifest.txt)

~/YOLO/

.
├── 000000000139.txt
├── 000000000285.txt
├── 000000000632.txt
├── 000000000724.txt
...
├── 000000581206.txt
├── 000000581317.txt
├── 000000581357.txt
├── 000000581482.txt
├── 000000581615.txt
└── 000000581781.txt

000000581781.txt

46 0.446 0.557 0.465 0.209
46 0.517 0.851 0.363 0.128
46 0.939 0.05 0.122 0.071
46 0.786 0.027 0.11 0.054
46 0.171 0.247 0.19 0.139
46 0.865 0.773 0.27 0.372
46 0.111 0.552 0.215 0.333
46 0.51 0.744 0.376 0.207
46 0.811 0.377 0.25 0.36
46 0.955 0.388 0.09 0.181
46 0.195 0.333 0.153 0.224
46 0.036 0.183 0.065 0.357
46 0.496 0.45 0.389 0.132
46 0.499 0.52 0.998 0.956

./manifest.txt

~/COCO/val2017/000000289343.jpg
~/COCO/val2017/000000061471.jpg
~/COCO/val2017/000000472375.jpg
~/COCO/val2017/000000520301.jpg
~/COCO/val2017/000000579321.jpg
~/COCO/val2017/000000494869.jpg
...
~/COCO/val2017/000000097585.jpg
~/COCO/val2017/000000429530.jpg
~/COCO/val2017/000000031749.jpg
~/COCO/val2017/000000284282.jpg

TODO

Refactoring (Release v2.0.0)

  • Add strict Type Annotation in code
  • Separate role in class more strictly
  • Rewrite README.md for more helpful use first
  • Resolve the problem that strictly validation check when the trivial error
  • Supported Multiprocessing
  • Skip object class that don't want