lukashaas/pigeon

Code for the CVPR 2024 paper highlight and demo "PIGEON: Predicting Image Geolocations".

pigeon's Issues

Few requests for experiments and citation

Hello. Thank you for your excellent work. I've been trying to cite your paper after conducting a few validations, but it seems the weights are not available, making it difficult to proceed. Could you please upload them? Additionally, I'm having trouble running the 'run.py' file because I can't locate the 'multiprocess.py' file. Would you be able to upload it as well?

Clarification of dataset splits

Thanks to authors for sharing their code.

Can the dataset splits be clarified? That is, can you provide the dataset(s) used for training and validation, as well as the specific split if training / validation are coming from the same dataset.

It is unclear what specifically comprises the validation dataset from the paper (e.g., whether it is a split of MP16, etc). This detail is important to ensure recreation of experiments and fair comparisons of future work.

The split seems to be coming from some metadata file according to the code here.

Thanks!

How to get admin_2.geojson？

Hi! Thank you and your amazing work.
When I want to run geocell_creation.py, I get an error: fiona._err.CPLE_OpenFailedError: data/geocells/admin_2.geojson: No such file or directory
Could you please explain how to get this file？

Pretraining question

Hi! very nice repository and paper. I have a question about the continued pretraining. Were all 428M parameters of OpenAI's CLIP ViT L/14 336 trainable for your continued pretraining? Or were some of them frozen?

Based on the paper, it seems like a batch size of 32 image/text pairs fit into 80GB of GPU memory, so I wanted to double check how many parameters of the original backbone were frozen/trainable. Thanks!

KeyError when run evaluate

Hello, amazing job, I trained on my own data and want to run an evaluate, but an error occurs when loading geocells:

lla_coords = torch.tensor(geo_df[['lng', 'lat']].values)
KeyError: "None of [Index(['lng', 'lat' ], dtype='object')] are in the [columns]"

I guess it is because the geocell generated by my run has only 6 columns: name, admin_1, country, size, num_polygons, geometry, I made my geocells.csv using dataset_creation/geocell/geocell_creation.py
So what might have gone wrong?

No such file

No such file or directory: 'data/driving_side/countries_driving_side.json'