Glass of Water

• Group Members: Alex Beeston, Malachi Harper, & Kaleb Lott
• Group Project Manager: Alex Beeston
• Topic of Research: Simulation of robotic manipulators that make decisions using computer vision.

Description

This program first classifies images of concrete based on whether or not they are cracked. Images of concrete, both cracked and intact, are in data/croppedImages/cracked and data/croppedImage/intact, respectively. The program src/main.py classifies these images using the ResNet18 neural network with 5-fold cross validation. With the default configurations (located in src/configurations.py), each of the five folds took about 20 minutes running on a CPU on Ubuntu 20.04. The program prints out the training and validation accuracies of the classifier after each epoch of each fold. The classifier achieves ~94% accuracy on the validation sets.

Instructions

1. If pipenv is not installed, install pipenv. On Windows (and probably Linux), should be just pip install pipenv.
2. Create the virtual environment with pipenv install.
3. Activate virtual environment with pipenv shell.
4. Run the classifier as python3 src/main.py dir/to/positive/samples dir/to/negative/samples. For example, to classify the sample images, run python3 src/main.py data/croppedImages/cracked data/croppedImages/intact.
5. Observe or redirect the console output to a log file.

If running the segmenter: run segmentor.m, rename the targetted file to whatever file the segmenter should be run on.