Query in correspondence of samples across datasets about ev2hands HOT 2 CLOSED

Hi @pmkalshetti,

I'm sorry for not getting back to you sooner. I did not receive notifications for Github.

For the datasets, the start_frame is synchronised across the event and RGB streams with their corresponding GTs. Hence,
subject_{id}_event.pickle) , MANO_Parameters (event_mano_params.pkl) , MANO_Parameters (rgb_mano_params.pkl), RGB_Data (subject_{id}_rgb.mp4) all have synchronised starting index.

You can clip the exceeding frames or the frame correspondences at the end.

To verify this, I ran this on the MANO meshes obtained by both the streams,

transformed_event_mesh = event_mesh.apply_transform(event_to_rgb_transform); print(np.abs(transformed_event_mesh.vertices * 1000 - rgb_mesh.vertices * 1000).mean())
The error comes out to be 0.023 mm (approx).

Also, I checked the sync scripts for RGB and event streams with the multi-view studio (to obtain the Captury GTs) and they too have synchronised starting index.

Best,
Christen

from ev2hands.

That's great; I'll clip the frame indices at the end. Thank you for verifying :)

from ev2hands.