Quantization method unweighted average returns NaN's

[maxim0815]

Describe the bug

NaN's appear after some time.

---

To Reproduce

I'm using the following to generate the sparse tensor out of some point cloud:

pts = x[:,:3]
feat = x [:,3:] / 255
in_field = ME.TensorField(
    features=feat,
    coordinates=ME.utils.batched_coordinates([pts / self.voxel_size], dtype=torch.float32),
    quantization_mode=ME.SparseTensorQuantizationMode.UNWEIGHTED_AVERAGE,
    minkowski_algorithm=ME.MinkowskiAlgorithm.SPEED_OPTIMIZED,
    device=self.device,
)
x = in_field.sparse()

Afterwards, I'm using the model for doing semantic predictions:

class MinkUNet34C(MinkUNet34):
    PLANES = (32, 64, 128, 256, 256, 128, 96, 96)

After some training steps the model returns NaN values.

Replacing: ME.SparseTensorQuantizationMode.UNWEIGHTED_AVERAGE with ME.SparseTensorQuantizationMode.RANDOM_SUBSAMPLE

Like in issue #273 resolves the problem. Nevertheless, I would be interested to know if it is really due to the QuantizationMode and why it does not appear from the beginning on. Or am I misunderstanding something else?

---

Expected behavior

Both QuantizationMode's should return values.

---

Desktop (please complete the following information):

• OS: Docker environment

• Python version: 3.8.10

• Pytorch version: 1.8.2

• CUDA version: 11.6

• NVIDIA Driver version: 510.47.03

• Minkowski Engine version 0.5.4

• ME.print_diagnostics(): ==========System========== Linux-5.10.0-1057-oem-x86_64-with-glibc2.29 DISTRIB_ID=Ubuntu DISTRIB_RELEASE=20.04 DISTRIB_CODENAME=focal DISTRIB_DESCRIPTION="Ubuntu 20.04.3 LTS" 3.8.10 (default, Nov 26 2021, 20:14:08) [GCC 9.3.0] ==========Pytorch========== 1.8.2 torch.cuda.is_available(): True ==========NVIDIA-SMI========== /usr/bin/nvidia-smi Driver Version 510.47.03 CUDA Version 11.6 VBIOS Version 94.02.42.00.2F Image Version G001.0000.03.03 GSP Firmware Version N/A ==========NVCC========== /usr/local/cuda/bin/nvcc nvcc: NVIDIA (R) Cuda compiler driver Copyright (c) 2005-2020 NVIDIA Corporation Built on Wed_Jul_22_19:09:09_PDT_2020 Cuda compilation tools, release 11.0, V11.0.221 Build cuda_11.0_bu.TC445_37.28845127_0 ==========CC========== /usr/bin/c++ c++ (Ubuntu 9.3.0-17ubuntu1~20.04) 9.3.0 Copyright (C) 2019 Free Software Foundation, Inc. This is free software; see the source for copying conditions. There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

       ==========MinkowskiEngine==========
       0.5.4
       MinkowskiEngine compiled with CUDA Support: True
       NVCC version MinkowskiEngine is compiled: 11000
       CUDART version MinkowskiEngine is compiled: 11000
  

---

Additional context

Thanks for sharing and supporting the code with the community. This project is really good.