GeoGuessr Location Predictor

A convolutional neural network trained on over 10,000 Google Street View images to predict the country depicted in a given image. The model provides a benchmark towards what types of AI models would work best with image-country tag recognition.

The dataset was collected via a customized HuggingFace dataset and later augmented with flips and color jitter. Various other methods for benchmarking included Log Scaling + K-Nearest Neighbors, but CNNs proved to be the most accurate.