ar-mine / IMG_Automator

GUI wrapper of U^2-Net for easier YCB like dataset making.

IMG_Automator

A GUI wrapper created by DearPyGui for pre-processing images to generate templates that are used as dataset.

Functions

• Use FFmpeg to separate videos to images for further processing.
• Use U^2-Net to make YCB like dataset.

Installation

The source codes have been tested with python3.8, and the packages needed are written into the requirements.txt.

You can just use git clone https://github.com/ar-mine/IMG_Automator.git to download the whole repo.

Usage

• Modify the main.py to specify use config-win.yaml or config.yaml and then run main.py.
  • It will load pre-trained U^2-Net model and then load config file if it exists.
  • Put your video that contains target objects in the path of video folder and then click separate button, then the video will be separated into sequence images according to the rate of slider.
  • Or you can just put all images in input folder and click process button, then it will generate images only contain target and save them in temp folder.
  • Finally, YCB-like images dataset will be saved in result folder with resolution of 120*120.
  • You can use open to see the images in the folder and use clean and related checkbox to clean redundant images or videos (the action will eraser the whole folder specified except the folder itself).
• The config-win.yaml and config.yaml store the default processing paths based on different formats of Windows and Linux. You can change it according to your project or modify the path while running the program.

Resource

Test video link: https://drive.google.com/file/d/17pofu-ii4KtUTX7YxbwOes9vlHXXils3/view?usp=sharing. By default, you can put the video in the IMG_Automator\dataset\video folder.