Why DER on Ego4D dataset is so bad ?

[amitli1]

According to: https://huggingface.co/pyannote/speaker-diarization,

the performacne of PyAnnote speaker diarization on Ego4D dataset is very bad (very high DER rate):

• Is there a reason for this bad results ?
• What is the different between Ego4D dataset and the other datasets that the performance is inferior ?