Why is the queue_length set to 5 and 3

[shenyinzhe]

Hello,

In the stage 1, queue_length = 5 In the stage 2, queue_length = 3 I'm wondering is there any specific reason to set it like above. Especially in the stage 2:

• Why is it reduced
• Why not follow nuscenes prediction challenge, where it utilizes 2 seconds past, i.e., 4 past timesteps.

Thanks

[YTEP-ZHI]

Hi @shenyinzhe, the queue length is reduced in the second stage for efficiency concerns, since a long queue of historical features is extremely memory-consuming.

[YTEP-ZHI]

The choice of queue_length=3 is more of an empirical practice, as there's no significant performance gap between queue_length 3 and 4. Therefore, we opt for queue_length=3 as it's more efficient in training.

[shenyinzhe]

Thanks for reply, I'll close the issue.