Tai Wang

Happened to see this issue :) What we used before is this [gt.bin](https://console.cloud.google.com/storage/browser/waymo_open_dataset_v_1_2_0/validation/ground_truth_objects;tab=objects?prefix=&forceOnObjectsSortingFiltering=false) version in Waymo 1.2.0. I observe that your mentioned new version of gt.bin is a little larger...

It may be related to the communication between your two machines. Please make sure they can successfully ping each other. Besides, you can also first try to 'train' a model...

Great! After a brief review of the difference of this repo with mmdet3d, I think the basic structure is ok. In my opinion, the main changes are focused on a...

We have just supported the occupancy visualization demo in #62 and the grounding one will be released with new methods and datasets.

I have also seldom encountered such cases recently, and maybe we will have a closer look at this issue. At the same time, we welcome more cues/information about this problem...

Thank @henryzhengr for providing a work-around solution. I close this issue for now and welcome further discussions if there are new observations.

I am sorry that the log is incomplete because the full dataset training is easy to interrupt before due to large computational burden. It's hard to find the original one,...

Do you use our officially provided data? It is really strange to allocate such huge amount of memory for the multi-view 3D detection model.

Maybe [EmbodiedScan](https://github.com/OpenRobotLab/EmbodiedScan) can meet your expectations.

You can try to follow [DORT](https://github.com/OpenRobotLab/DORT) to produce 3D tracklets from 3D detection results. The open-source version is based on BEV-based detectors, and we also have preliminary attempts based on...