search

sharathadavanne / seld-dcase2022

Baseline method for sound event localization task of DCASE 2022 challenge
50 stars 21 forks source link

It seems there some issues in SELD_evaluation_metrics.py #2

Closed Jinbo-Hu closed 2 years ago

Jinbo-Hu commented 2 years ago

Hi, It seems there are some issues in SELD_evaluation_metrics.py. Line 137: gt_doas = gt_arr[:, :-1] and Line 141 pred_doas = pred_arr[:, :-1]. It seems that gt_arr[:, 0] and pred_arr[:, 0] are source number indexes, rather than gt_arr[:, -1] and pred_arr[:, -1].

sharathadavanne commented 2 years ago

Thanks for participating in the challenge @Jinbo-Hu

The variables gt_arr and pred_arr are of the form [azimuth_angle_degrees, elevation_angle_degrees, track_id] So using gt_arr[:, :-1] and pred_arr[:, :-1] should ideally give us [azimuth_angle_degrees, elevation_angle_degrees] I checked it again on my end. Could you verify the same from your side?

Jinbo-Hu commented 2 years ago

Thanks for your quick response!

I run your baseline code. Here is one of outputs of gt_arr, gt_doas, pred_arr, pred_doas

gt_arr: [[ 1. 35. -12.]] gt_doas: [[ 1. 35.]] pred_arr: [[ 0. -101.32076171 -15.99277551]] pred_doas: [[ 0. -101.32076171]]

It still seems that gt_arr[:, 0] and pred_arr[:, 0] are source number indexes.

sharathadavanne commented 2 years ago

You were right @Jinbo-Hu Thanks for spotting the bug. We have now fixed the code. Check this commit to see the difference in the SELD_evaluation_metrics.py file