Understanding use of pretrained models

[brentgti]

Hi @Sachini,

I am trying to use RoNIN to deduce location where GNSS satellite signals are poor or unavailable (like in urban canyons, not indoors). What I want to be able to do is deduce location when starting from a known location (where GNSS signal is available) and moving to where location can't be determined using GNSS.

I've read other posts (including #20, #25, and #41) here about using the pretrained models. I've gotten IMU data from an Android device using the app you provided and was able to run compile_dataset_h5.py to generate data.hdf5 and info.json files with the --unannotated switch, but the info.json file seems to be incomplete, compared to the info.json files I see in the datasets you provide.

When I try to run ronin_resnet.py, I get KeyError: 'gyro_integration_error' in select_orientation_source in data_utils.py. I assume that this is because I'm not properly incorporating data from the game rotation vector (grv), but I don't understand how to do that. Can you provide more guidance on that? Thanks in advance.