/Making_Dataset-For-Yolov3v4

COCO, VOC ,dataset , Ultralytics version - yolov3v4

Primary LanguagePython

Make_Dataset-for-Ultralytics-yolov3v4:

We make the dataset for ultralytics-version yolo (https://github.com/ultralytics/yolov3 )

And the origianl darknet-version yolo (https://github.com/AlexeyAB/darknet ),

Both support for yolov3 and yolov4.

This repository mainly include three parts:

  • Part I: make an environment for the Darknet

  • Part II: make_VOC_dataset

  • Part III: make_COCO_dataset and test on the test2017.

Part I: make an environment for the Darknet

  • Pelease read about the ./Env_For_Darknet.md

Part II: make_VOC_dataset

  1. Make_VOC_dataset : this three python files realized the function of convert the label from the ".xml" format to the ".txt" format, so that we can train the YOLO with the .txt format label.
  • If we make it out, the VOC dataset in the data folder should be the following hierarchy.
├── data
│   ├── VOC         For  VOC dataset  folder
│   |     ├── images
|   |     |        ├── train
|   |     |        ├── val
|   |     ├── labels
|   |     |        ├── train
|   |     |        ├── val
|

Part III: make_COCO_dataset and evaluate on the test2017

3.1 make_COCO_dataset:

The first three python files realized the function of convert the label from the " .json" format to the ".txt" format, so that we can train the YOLO network with the .txt format label.

-If we make it out, the COCO dataset in the data folder should be the following hierarchy.

├── data
│   ├── COCO      For  COCO dataset folder 
│   |     ├── images
|   |     |        ├── train
|   |     |        ├── val
|   |     ├── labels
|   |     |        ├── train
|   |     |        ├── val
|

3.2 If you want test your model without the ground truth label.

And submit the reuslts.json to the CodaLab https://competitions.codalab.org/competitions/20794 ,

Here we have a simple tutorials: ./make_COCO_dataset/readme_test-dev2017.md