fix bug to allow pointnet to train in fp16

[maskjp]

Thanks for the great repository!

Motivation

I try to train PointNet in fp16 mode but the function/modules don't allow HALF input, so I made some changes.

Modification

• I changed '''gather_points''' ops to allow Half input because this operation is just indexing corresponding values. And I add two lines in the '''tests/test_models/test_common_modules/test_pointnet_ops.py''' file.
• I add '''force_fp32''' in the '''mmdet3d/ops/group_points/group_points.py''', because it relates to knn or ball_query operations.

BC-breaking (Optional)

Does the modification introduce changes that break the back-compatibility of the downstream repos?
If so, please describe how it breaks the compatibility and how the downstream projects should modify their code to keep compatibility with this PR.

Use cases (Optional)

If this PR introduces a new feature, it is better to list some use cases here, and update the documentation.

Checklist

1. Pre-commit or other linting tools are used to fix the potential lint issues.
2. The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.
3. If the modification has potential influence on downstream projects, this PR should be tested with downstream projects.
4. The documentation has been modified accordingly, like docstring or example tutorials.

[CLAassistant]

All committers have signed the CLA.

[ZwwWayne]

Hi @maskjp ,
Thanks for a great job! Could you please sign the CLA so that we can merge this PR eventually? Check the box below:

Furthermore, could you run the unit tests in your local machine with GPU to make sure the operator works correctly?

[ZwwWayne]

We are migrating CUDA ops to MMCV. There is already a refined version of our CUDA ops in MMCV. If you are interested, you can also create a PR to support FP16 training of PointNet ops in https://github.com/open-mmlab/mmcv/tree/master/mmcv/ops/csrc/pytorch. If you are not interested, do not worry, the team will follow up and update the MMCV ops.

[maskjp]

Hi @maskjp , Thanks for a great job! Could you please sign the CLA so that we can merge this PR eventually? Check the box below:

Furthermore, could you run the unit tests in your local machine with GPU to make sure the operator works correctly?

Hi,

I signed the agreement using this account before. But it seems Github automatically added another my account as a committer. Just solve it.

Yes, I ran the test on my machine with GPU, and they worked under the fp16 mode.

Thank you!

[maskjp]

We are migrating CUDA ops to MMCV. There is already a refined version of our CUDA ops in MMCV. If you are interested, you can also create a PR to support FP16 training of PointNet ops in https://github.com/open-mmlab/mmcv/tree/master/mmcv/ops/csrc/pytorch. If you are not interested, do not worry, the team will follow up and update the MMCV ops.

Thanks for this great work! I will take a look at it.

[maskjp]

Hi, @ZwwWayne,

There's an inconsistency of the visions of isort between online check and the pre-commit file. The online one used v5.0.0 and the pre-commit used v5.0.2.

v5.0.0

import torch
from mmcv.runner import force_fp32
from torch import nn as nn
from torch.autograd import Function
from typing import Tuple

from ..ball_query import ball_query
from ..knn import knn

v5.0.2

from typing import Tuple

import torch
from mmcv.runner import force_fp32
from torch import nn as nn
from torch.autograd import Function

from ..ball_query import ball_query
from ..knn import knn

[OpenMMLab-Assistant003]

Hi @maskjp！First of all, we want to express our gratitude for your significant PR in the MMDetection3D project. Your contribution is highly appreciated, and we are grateful for your efforts in helping improve this open-source project during your personal time. We believe that many developers will benefit from your PR.

We would also like to invite you to join our Special Interest Group (SIG) private channel on Discord, where you can share your experiences, ideas, and build connections with like-minded peers. To join the SIG channel, simply message moderator— OpenMMLab on Discord or briefly share your open-source contributions in the #introductions channel and we will assist you. Look forward to seeing you there! Join us ：https://discord.gg/UjgXkPWNqA

If you have WeChat account，welcome to join our community on WeChat. You can add our assistant ：openmmlabwx. Please add "mmsig + Github ID" as a remark when adding friends：）
Thank you again for your contribution❤ @maskjp