no valid points: scale 0

[vvcatstar]

When I run the example data , the error is no valid points: scale 0. My environment is

• RTX3090, torch 1.8+cu111, cuda11.1. (I try use the torch 1.6.0, but RTX3090 not match it. )
• The torchsparse is install by rebuild.
• repo commit is cd047e2356f68f60adfb923f33572d712f64b57a (HEAD -> master, origin/master, origin/HEAD)

The run code is python3 demo.py --cfg ./config/demo.yaml The log is:

2022-02-07 16:15:15.152 | INFO     | __main__:<module>:35 - Found SyncedPoses.txt, skipping data pre-processing...
2022-02-07 16:15:15.152 | INFO     | __main__:<module>:37 - Running NeuralRecon...
2022-02-07 16:15:15.154 | INFO     | __main__:<module>:50 - Initializing the model on GPU...
2022-02-07 16:15:17.009 | INFO     | __main__:<module>:58 - Resuming from ./checkpoints/model_000047.ckpt
2022-02-07 16:15:17.104 | INFO     | __main__:<module>:64 - Start inference..
  0%|                                                                                                                                                                           | 0/110 [00:00<?, ?it/s]2022-02-07 16:15:17.527 | WARNING  | models.neucon_network:forward:184 - no valid points: scale 0
  1%|█▍                                                                                                                                                                 | 1/110 [00:00<00:46,  2.32it/s]2022-02-07 16:15:17.590 | WARNING  | models.neucon_network:forward:184 - no valid points: scale 0
2022-02-07 16:15:17.654 | WARNING  | models.neucon_network:forward:184 - no valid points: scale 0
  3%|████▍                                                                                                                                                              | 3/110 [00:00<00:16,  6.30it/s]

[Cli98]

@vvcatstar I run into same issue. Do you find any solution?

[linzhenyuyuchen]

+1

[Cli98]

I carefully check the source code. This is due to torchsparse package. Under current setup, neuralrecon with torchsparse dependency can only work on cuda 10.2 and 11.1 or cpu mode.

[harsh-sculptsoft]

@vvcatstar based on https://github.com/mit-han-lab/torchsparse/blob/HEAD/docs/FAQ.md you can setup and it will run; it's worked for me