l5kit getting stuck during load_agents_mask about l5kit HOT 7 CLOSED

Which set are you working with? That function simply loads the mask from the disk, it should be quite fast. Did you generate the agent_mask correctly on the first run?

from l5kit.

i'm using full train dataset. I didn't generate agent_mask.

Seems that it happens in case more than one pipeline (not necessarily l5kit) are running.

from l5kit.

pipeline?

from l5kit.

I've met the same problem. Using zarr.open_array("./agents_mask/0.5") can handle this. (too many tiny files)

from l5kit.

pipeline i mean another pytorch lightning train loop

from l5kit.

pipeline i mean another pytorch lightning train loop

This has never been tested with PTL, so I can't guarantee it works there :(

from l5kit.

I'm closing this for now as we don't officially support PTL (yet)

from l5kit.