/CorrosionDetector

Corrosion Detection from Images

Primary LanguageJupyter Notebook

CorrosionDetector


This notebook uses Transfer Learning on VGG16 Convnet, for the purpose of identifying images with rust. Achieves 87% F1-score by training on images downloaded from internet. Can correctly identify reddish brown painted images as no-rust.
The second part contains the notebook used to localize Rust using Tensorflow Object Detection API

Run on FloydHub