Latitude Attitude predicts geographic coordinates based solely on image features using deep learning. It uses a ResNet model with dropout and batch normalization, trained on over 3,600 samples with 150,000 input features, reaching a 34.6-meter RMSE on the test set. Baseline models using Linear Regression and Vision Transformers were also developed, and the project placed 2nd out of 200+ students.
Latitude Attitude predicts geographic coordinates based solely on image features using deep learning. It uses a ResNet model with dropout and batch normalization, trained on over 3,600 samples with 150,000 input features, reaching a 34.6-meter RMSE on the test set. Baseline models using Linear Regression and Vision Transformers were also developed, and the project placed 2nd out of 200+ students.
