Code for recreating the results in our ICCV 2019 paper.

demo.py is a simple demo script that either 1) takes location as input and returns a prediction for all the categories predicted to be present at that location or 2) generates a dense prediction for a category of interest.
geo_prior/ contains the main code for training and evaluating models.
gen_figs/ contains scripts to recreate the plots in the paper.
pre_process/ contains scripts for training image classifiers and saving features/predictions.
web_app/ contains code for running a web based visualization of the model predictions.

For more results, data, and an interactive demo please consult our project website.

If you find our work useful in your research please consider citing our paper.

@inproceedings{geo_priors_iccv19,
  title     = {{Presence-Only Geographical Priors for Fine-Grained Image Classification}},
  author    = {Mac Aodha, Oisin and Cole, Elijah and Perona, Pietro},
  booktitle = {ICCV},
  year = {2019}
}