CS230-Microcrystal-Facet-Segmentation

Microcrystal facet segmentation algorithm based on U-NET architecture.

Table of contents

• General info
• Screenshots
• Technologies
• Setup
• Features
• Status
• Inspiration
• Contact

General info

The goal of this project is to train an appropriate CNN architecture that is able to perform semantic segmentation of cuprous oxide Cu2O nanocrystal facets.

Baseline Model

keras implementation (https://github.com/divamgupta/image-segmentation-keras/)

Screenshots

Technologies

• python - version 3.6.5
• keras - version 2.3.0
• keras_segmentation
• opencv_python - version 4.2.0.32
• Augmentor - version 0.2.8

Code Examples

Show examples of usage:

from keras_segmentation.models.unet import unet_mini

model = unet_mini(n_classes=4,  input_height=96, input_width=96  )

model.train(
    train_images = "Dataset/train/",
    train_annotations = "Dataset/train_labels/",
    checkpoints_path = "Dataset/checkpoints",
    val_images = "Dataset/test/",
    val_annotations = "Dataset/test_labels/",
    epochs=50, validate=True, batch_size=8, 
    optimizer_name="adam",
    gen_use_multiprocessing=True,
    auto_resume_checkpoint=False,
    val_batch_size=2,
)

Features

List of features ready and TODOs for future development

• Train on 3 different U-NET architecture variants

Status

Project is: finished

Report

CS230 Winter 2020 (http://cs230.stanford.edu/projects_winter_2020/reports/32641590.pdf)