Seeking suggestions for evaluating the Ego4D dataset

[sdqdlgj]

Hi Xizi,

Thanks for open-sourcing the LoCoNet code, which demonstrates excellent performance.

May I inquire about how to evaluate the mAP performance of the Ego4D dataset?
The official Ego4D repository provides mAP evaluation code (https://github.com/EGO4D/audio-visual/tree/main/active-speaker-detection/active_speaker), but I have a few questions:

1. The official Ego4D repository presents a complex testing process, and the format of some files, such as val.txt, v.txt and etc., are unclear. Do you have any usage suggestions for the Ego4D evaluation code?
2. I would like to know whether the mAP metric for Ego4D is calculated in the same way as the mAP metric for the AVA dataset?

Thank you in advance!

Best,
Guanjun