Tai-Wang

icon indicating a website link https://tai-wang.github.io/ icon indicating an email [email protected]

icon company Shanghai AI Lab icon location Shanghai icon location Researcher at OpenRobotLab, Shanghai AI Lab, PhD from MMLab, CUHK.

Results 98 comments of Tai-Wang

Do benchmark.py and get_flops.py work for point cloud detection models too ?

Yes, you can have a try. Please let us know if there is any problem.

Can more detectors be trained with SUN-RGBD?

Currently no more, because there are still gaps between different datasets and model hyperparameters. I would encourage you to have some attempts by yourself if you have some experiences of...

AssertionError: StepLrUpdaterHook: "warmup_iters" must be a positive integer

This is not a typical format for mmdet3d config. Do you add new configs by yourself? Please show your config and provide more information.

Got size error after update waymo datasets whth the script update_data_coors.py

Yes, please refer to the [compatibility doc](https://mmdetection3d.readthedocs.io/en/latest/compatibility.html) for more details.

The use of ObjectSample

It's just a kind of data augmentation. The usage of it sometimes may bring performance gains while sometimes not. We usually add it when it brings some gains. Besides, we...

waymo convert about CALIB file in kitti format！

Could you please describe this issue more clearly with some references to specific codes?

Dataloader blocked using distributed training

It's hard to say what is the problem exactly on my side. Maybe this is related to the IO/CPU setting or your machine. For example, your machine may not be...

Lyft Test Fail with KeyErr

We do not support centerpoint on the Lyft for now. Do you modify the config by yourself to make it compatible with Lyft?

When I use the pointpillar model to infer my own point cloud data, there are errors. I don't know why.

I don't see any error from the provided messages. All are warnings and you can just ignore them.

Why not stack the tensors(like point clouds) along batch-dimension?

Actually, this is overall a trade-off of efficiency and memory overhead. The padding operation will sacrifice more memory to save some padded input while the list manner can not avoid...