Crack-segmentation icon indicating copy to clipboard operation
Crack-segmentation copied to clipboard

Published 20 hours ago • verified publisher icon choiw-public
→

Metadata

This repo contains customized deep learning models for segmenting cracks.

Crack Segmentation

A deep learning model for segmenting cracks. This repository will have more models in future.

Demo

  • Full demo: https://bit.ly/choiw-fulldemo

  • Crack-propagation1:
    alt text

  • Crack-propagation2:
    alt text

  • Crack-moving shot1:
    alt text

  • Crack-moving shot2:
    alt text

  • Asphalt-moving shot:
    alt text

Libraries

  • Python v3.x
  • TensorFlow v1.15
  • TensorFlow-Probability 0.7.0
  • Horovod (https://github.com/horovod/horovod)
  • OpenCV
  • NumPy
  • SciPy

Key techniques

  • Half-precision (FP16)
  • Feature pyramid
  • Global context block
  • Bottleneck layer
  • Shortcut connection (concatenation and addition)

Model summary

  • Dataset: built from scratch for this side project
  • Architecture: depth-42 (38 downscale and 4 upscale), no pretrained model used
  • Number of parameters: 2 million

← Metadata

49
Stars
7
Forks
Watchers

Owner

verified publisher icon choiw-public

Metadata

This repo contains customized deep learning models for segmenting cracks.

Back