Hugues THOMAS

Well, this seems pretty clear, your labels seem to have float64 dtype instead of integer. You should be able to solve this yourself by following the code where the labels...

Sure, you just have to change the number of features to 1 in the configuration here: https://github.com/HuguesTHOMAS/KPConv-PyTorch/blob/e600c1667d085aeb5cf89d8dbe5a97aad4270d88/train_S3DIS.py#L146 If you want more control on the input features you want to add,...

the features do not include x, y, z. For S3DIS they are: 1, r, g, b, h, where 1 is just a column of 1, and h is the height...

Hi @cystal-lili, Your modification is good and you should not modify anything in `tester.py`. But make sure you used this modification also during training, otherwise, your results will not be...

I am sorry I cannot help you with windows issues. I only run the code on Linux myself. I advise you to search for this error on google and if...

Hi @liangzhendong123, If your 2D point clouds are sparse, using KPConv on your 2D data could make sense. Theoretically, it is totally possible, that we can define a point kernel...

I have other projects right now, but I see that many people asked for it, so I will try to take some time to do it in the next weeks.

@P-Light @lqzhao @LeopoldACC, I just added [pretrained weights for S3DIS](https://github.com/HuguesTHOMAS/KPConv-PyTorch/blob/master/doc/pretrained_models_guide.md). some of you still need them, I'd be happy to know if the instructions I gave work properly. Best, Hugues

Hi @luzonghao1, The reason the memory changes is because a lot of dimensions are variable, mainly the number of points in each batch. You can change a simple line so...

The error is probably here: https://github.com/HuguesTHOMAS/KPConv-PyTorch/blob/3a774ff8d54a4d080fe65093b2299ede35d9735d/utils/tester.py#L433-L434 maybe change with: ```python prob_names = ['_'.join(test_loader.dataset.label_to_names[label].split()) for label in test_loader.dataset.label_values if label not in test_loader.dataset.ignored_labels] ```