Nireil

> I had the same problem，tried pytorch 1.8.1 I have solved it! You need to add `from open3d.visualization.tensorboard_plugin import summary` **before** `from open3d.visualization.tensorboard_plugin.util import to_dict_batch` in the /examples/visualization/tensorboard_pytorch.py.

i met similar question, did you solve it? when i run python main_SensatUrban.py --mode test --gpu 0, the progress was killed as follows no matter using tensorflow1.11&cuda9.0 or tf1.14&cuda10.0. ...

> i met similar question, did you solve it? when i run python main_SensatUrban.py --mode test --gpu 0, the progress was killed as follows no matter using tensorflow1.11&cuda9.0 or tf1.14&cuda10.0....

Would you mind share me the iou and weight file you test out on toronto3d?

How about the progress of the check? I wonder if you add the intensity to the feature, but when I add the intensity feature, the loss is NAN.

> > > > > Intuitively， without rgb, road_marking is almost impossible to be segmented correctly like other model (kpconv, TGNet), because there are few differences between road and road_marking...

> I have tried to generate experimental code on this basis, but perhaps because of my personal ability, I failed. I wonder if you would mind publishing the relevant code?...

> 作者您好，请问能否提供一下toronto-3d数据集训练randlanet的预训练模型吗 我的邮箱[[email protected]](mailto:[email protected]) 请问你收到这个RandLA-Net在这个数据集上的权重了吗？我自己训练一直达不到这么高的miou，用上rgb也只有70出头

> 作者大大，请问指标中的OA计算时分母算上被忽略的类别了吗？

And I think that the labeled result of 'colors.xml' is clearer than the paper's figure : )