UC_Merced_classification

Mobile net based classification for UC merced data

Accuracy metrics

Accuracy: 0.97

Model Outputs

Getting data

Go to http://weegee.vision.ucmerced.edu/datasets/landuse.html

Data Directory

Create a new folder named data with with images in folder structure (.// data // UCMerced_LandUse // UCMerced_LandUse// Images //* // *.tif)

File Structure

The helper folder contains io_tools and viz_tools.

1. The io_tools subsets the data to get 14 classes and subset the data into training validation and testing in 70:10:20 ratio. The io_tools also has a function thatloads data as normalized array
2. Viz_tools helps exploratory data analysis and also helps plot misclassified images to identify algorithm efficacy