VoxelNeXt icon indicating copy to clipboard operation
VoxelNeXt copied to clipboard

ONNX export

Open yBeOne opened this issue 2 years ago • 5 comments

RuntimeError: Only tuples, lists and Variables are supported as JIT inputs/outputs. Dictionaries and strings are also accepted, but their usage is not recommended. Here, received an input of unsupported type: numpy.ndarray. Have you export VoxelNeXt to ONNX?

yBeOne avatar Jul 04 '23 13:07 yBeOne

Hi,

I think this might be hard. Because the spconv backbone receives sparse tensors, which is not supported yet by ONNX.

I have converted the head of VoxelNeXt to ONNX. That is easy. Please refer to this one.

Regards, Yukang Chen

yukang2017 avatar Jul 05 '23 04:07 yukang2017

Hi, There is another question that how to count the distribution of query voxels like Tabel 7.

yBeOne avatar Jul 10 '23 12:07 yBeOne

Hi,

I think this might be hard. Because the spconv backbone receives sparse tensors, which is not supported yet by ONNX.

I have converted the head of VoxelNeXt to ONNX. That is easy. Please refer to this one.

Regards, Yukang Chen

Hi, Dear Yukang

can you share where is the link or postion for "this one"?

Thanks!

HuangVictorAuto avatar Jul 13 '23 03:07 HuangVictorAuto

I think this might be hard. Because the spconv backbone receives sparse tensors, which is not supported yet by ONNX.

Could it be done via this plugin?

https://github.com/jingyue202205/SE-SSD-AI-TRT/blob/master/sparseConv3dlayer.h

ArseniuML avatar Sep 27 '23 14:09 ArseniuML

Hi,

I think this might be hard. Because the spconv backbone receives sparse tensors, which is not supported yet by ONNX.

I have converted the head of VoxelNeXt to ONNX. That is easy. Please refer to this one.

Regards, Yukang Chen

Could you please provide config to test ONNX model?

Gaussiandra avatar Apr 19 '24 19:04 Gaussiandra