Using intensity as a feature in training results in gradient vanishing

[RBinSky]

Hi THOMAS, your work is great, but currently I have a problem. We used the WHU Urban3D dataset as input, which contains only xyz and intensity , so I used the intensity as feature input, and before training the network, we performed a normalisation operation on the intensity to a range between 0-1, but during the training period, we noticed that the gradient disappears not long after the network has been trained. Has this ever happened to you before? Thank you very much!