Loic Landrieu

S3DIS is composed of 6 levels across 3 buildings, referred to as *folds*. They are classically trained by cross-validation: we select a fold for testing and train on the rest,...

Hi, sorry I missed your issue. `components[i_com]` is the content of the superpoints `i_com`, ie the list of indices of the points it contains. Could you print the shapeand first...

The ground truth of semantic 3D is sparse, and only a heavily subsampled subset of points are annotated. Depending on the resolution and the voxelisation it can be more or...

I haven't coded the ground truth upsampling yet. It would look like something like that in visualize.py: ``` if gt_out and bool(args.upsample): if args.dataset=='s3dis': #TODO elif args.dataset=='sema3d': data_file = data_folder...

Hi, ver_batch is the number of lines to be read simultaneously. It should be around 5.000.000 rather than 4! In case this doesn't fix your issue, could printthe following sizes:...

Hi, sorry for delay. It is not provided in this repo because it is quite different from SPG. I have my own implementation in a private repo, which is nolonger...

I invited both of you to the private repo. The relevant code is in: `embedding/cluster_data_loader.py` and `embedding/cluster_embed.py`. If I remember correctly, the first function when called will create an "efficient"...

`python learning/evaluate.py --dataset vkitti --odir results/vkitti/best --cvfold 123456` should give you a breakdown of OA and IoU by classes.

Thanks for the tip, I'll look into it.

Thanks, that's useful. Adding it to the pile for the 0.2 version.