Hugues THOMAS

> In such cases is it acceptable if this class is excluded in the calculation of overall mean Iou? No, it is not, because you are not supposed to know...

I personally do not use the code on windows, so I am afraid I won't be of much help. This repo is aimed at Linux primarily, if you want to...

> have you considered adding a weak supervision module to the kpconv framework? No, this repo is really for defining a 3D point cloud convolution. It should serve as a...

I am not sure that I ever tried the visualization code on SemanticKitti. Visualization is an advanced feature so I did not spend time making sure it works in every...

Hi Eric, You can view the process of predicting Area4 as equivalent to segmenting an image that would have 1000000 x 1000000 pixels. As you can understand, that would take...

> Is there a specific reason why you set a normal distribution of N(-radius0, 2*radius0) for the sampling of random kernel points? Well, it is not a normal distribution, it...

Now, I tried this with the following involution parameters : ``` channels_per_group = 16 reduc_ratio = 4 ``` and I noticed: - the involution drastically reduces the number of parameters...

I didn't experiment a lot with SemanticKitti actually. I only tried 3 and 5 and did not optimize all parameters. 5 was slightly better, but also slower and more memory-consuming.

No, on the leaderboard, I was using `n_frames=3`

You can try to reduce num_votes to a lower value like 5 or 10: https://github.com/HuguesTHOMAS/KPConv-PyTorch/blob/8ec7478cd6b5fa3c6f423b984d6817b5e90e057b/utils/tester.py#L176