search

PJLab-ADG / DetZero

[ICCV 2023] DetZero: Rethinking Offboard 3D Object Detection with Long-term Sequential Point Clouds
https://superkoma.github.io/detzero-page/
Apache License 2.0
332 stars 32 forks source link

Daemon / Refine for training set #45

Closed Annie990401 closed 8 months ago

Annie990401 commented 10 months ago

I split train set in daemon, which PATH_TO_TRACKING_RESULT should be use. Because in tracking and detection part, the code only generate the val result. Should I generate result for training set?.

cd DetZero/daemon && python prepare_object_data.py --track_data_path --split train

superkoma commented 8 months ago

Absolutely yes, you need to run the detection and tracking for train set, and then prepare the object data for training. The command is right, you need to set --split train.

Isaac-xie commented 6 months ago

sorry, does this mean we need to use waymo's training set in the detection and tracking part?

YoushaaMurhij commented 2 months ago

@superkoma, @Isaac-xie, @Annie990401. I tried to run the detection and tracking for train set but I ran out of RAM. How much RAM do I need to prepare the data for train split? and can I reduce the required memory usage for this step? Thanks!