Support training of Voxel Mamba on Waymo

[gwenzhang]

Motivation

Support Voxel Mamba (https://arxiv.org/abs/2406.10700) in Openpcdet.

Modification

Add file pcdet/models/backbones_3d/voxel_mamba.py, which includes the sparse backbone of Voxel Mamba.
Add file pcdet/models/model_utils/voxel_mamba_utils.py, which includes the utility functions for Voxel Mamba.
Add file tools/process_tools/create_hilbert_curve.py, which includes the template codes for Voxel Mamba.
Change file pcdet/models/backbones_3d/__init__.py, which adds the Voxel Mamba in the init file.

Mamba

Download the mamba package from https://github.com/gwenzhang/Voxel-Mamba/tree/master/mamba or google drive (https://drive.google.com/file/d/12BfIyvsA_BiX3IzgTEFfhd6RCEigQPvW/view?usp=drive_link) and pip install -e . in local file for the reason (#960).

[Petros626]

If you work with Kaggle the following could be interesting for others:

1. !sed -i '107a\ cc_flag.append("-gencode")\n cc_flag.append("arch=compute_60,code=sm_60")' /kaggle/working/Voxel-Mamba/mamba/setup.py

2. !grep -n "cc_flag.append\|compute_6" /kaggle/working/Voxel-Mamba/mamba/setup.py

3. %cd /kaggle/working

4. !git clone https://github.com/Dao-AILab/causal-conv1d.git

5. !grep -n "cc_flag.append" /kaggle/working/causal-conv1d/setup.py

6. %cd /kaggle/working/causal-conv1d

7. !CAUSAL_CONV1D_FORCE_BUILD=TRUE pip install . --no-build-isolation

8. %cd /kaggle/working/Voxel-Mamba/mamba

9. !MAMBA_FORCE_BUILD=TRUE pip install . --no-build-isolation or in case of an error !MAMBA_FORCE_BUILD=TRUE pip install . --no-build-isolation --no-deps

Test the installation:

import torch
from causal_conv1d import causal_conv1d_fn
from mamba_ssm import Mamba
print("causal_conv1d & mamba_ssm succesfully imported")