Skip to content

Aoobi/train_unet

Repository files navigation

Unet training and inference repo

Repository to keep a working version of Unet training as well as inference and post-processing including analyzing results based on IEC standards.

Requirements and installation

  • Python 3.8

  • Install packages using provided requirements.txt file

pip install -r requirements.txt

Usage

Training

TBD

Inference

Batch inference on a folder

Use main_folder_inference.py and change the path to the folder containing V&G images. Standard operation involves locating fiber center and cuting the image to 768x768 - if images are already cut to required size, set full_size = False.

Chose desired IEC standard, all available standards are in the IEC_standards folder. Set standard name in the spec_path variable.

from yolov8_locate_fiber_and_cut import locate_cut_main
from unet_inference import unet_inference_folder
from IEC_watershed_welz_circle import analyze_to_IEC

if __name__ == '__main__':

    #select proper spec you want to evaluate against
    spec_path = r"IEC_standards/spec_IEC_ed2_SMPC_RL26dB.json"

    full_size = True # False if your images are already 768x768
    images_path = "path_to_your_folder"
    if full_size:
        locate_cut_main(images_path)

    unet_inference_folder(images_path)

    analyze_to_IEC(images_path, images_path, spec_path)

locate_cut_main currently overwrites the full size images in the folder

If locate_cut_main does not find a fiber in the image it will move the image to a new subfolder fiber_not_detected that will not be analyzed.

unet_inference_folder will load the cut images and for each image save a mask with name imageName_mask.png.

analyze_to_IEC loads masks and original images and returns one .png file with marked detections and one .txt file with counts of detections based on selected IEC standard.

Full-size image Cropped image Mask
uncut cut mask
Result with marked detections (IEC)
marked

Task list

  • Modify behaviour of locate_cut_main to not overwrite images, but create cut images in new folder instead
  • Add single image inference
  • Improve IEC defect count output formating
  • Add "Pass/Fail" information to the IEC output file (currently analyzed in code but not reported)

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages