Luc Hayward
Luc Hayward
This is the github for the original work: [Github](https://github.com/charlesq34/pointnet2)
It would be best to explain for other users and/or close this and open a new issue I think. You can use time.time() to get the current time and then...
Yes it does work for arbitrary class numbers (though these must exist in your training data). You need to change the final layers output size in the model definition.
See the comment in #72 [here](https://github.com/HuguesTHOMAS/KPConv-PyTorch/issues/72#issuecomment-771066086). My understanding is without the ones-vector the network will treat those points as "empty space". The ones-vector ensures that at minimum the local structure...
pptk is only supported up to python 3.7. You would need to recompile it yourself for newer versions or try and install the old version (last time I tried this...
There's always the possibility of recompiling the wheel yourself, I doubt there's much that will break. On Wed, 01 Jun 2022 at 19:02, garynobles ***@***.***> wrote: > Are there any...
I'm afraid that was just the suggestion given to me by another friend of mine. I love pptk for its simplicity and low overhead but Open3D may help you.
I would make sure that your data is actually being treated the same as the benchmark datasets and you end up with the same shape going into the model.
It takes about 24 hours on an RTX3080
I think this might be better as an example file rather than all written out into the readme?