search

nianticlabs / marepo

[CVPR 2024 Highlight] Map-Relative Pose Regression for Visual Re-Localization
https://nianticlabs.github.io/marepo/
Other
104 stars 8 forks source link

how to use #2

Closed xinyeDai closed 2 months ago

xinyeDai commented 2 months ago

How to use this after I download the datasets.

tcavallari commented 2 months ago

Hello, the main README file contains instructions on how to evaluate the pretrained marepo models (to replicate the results from the paper) or retrain a new marepo network from scratch.

For example, to evaluate the existing models on the various datasets: https://github.com/nianticlabs/marepo?tab=readme-ov-file#marepo-evaluation

Please let us know if you have further questions.

Qhijane commented 2 months ago

Hello, the main README file contains instructions on how to evaluate the pretrained marepo models (to replicate the results from the paper) or retrain a new marepo network from scratch.

For example, to evaluate the existing models on the various datasets: https://github.com/nianticlabs/marepo?tab=readme-ov-file#marepo-evaluation

Please let us know if you have further questions.

Hello, if I want to use your method to test my own dataset or RIO10 dataset, how can I replace your dataset with my own for testing?

tcavallari commented 1 month ago

Hi, to test on a different dataset you would have to create your own Dataset class.

You can look at dataset.py for details on the interface of the __getitem__ call (as usual for PyTorch datasets). Once you have implemented the new dataset class, then you need to update the testing script to actually use it, then you can call it from a bash script, like the one we used to run the tests on the 7Scenes dataset, here.