search

drprojects / superpoint_transformer

Official PyTorch implementation of Superpoint Transformer introduced in [ICCV'23] "Efficient 3D Semantic Segmentation with Superpoint Transformer" and SuperCluster introduced in [3DV'24 Oral] "Scalable 3D Panoptic Segmentation As Superpoint Graph Clustering"
MIT License
508 stars 65 forks source link

panoptic results #87

Closed meehirmhatrepy closed 3 months ago

meehirmhatrepy commented 3 months ago

after getting results for panoptic segmentation using code below, how do i print or save information in format like [label, instance, [x,y,z ]] for all instances predicted. also, panoptic predictions for kitti350 only show object and stuff and not label for that object. I am facing difficulty understanding that. please if possible explain that too. if task_widget.value == 'panoptic': raw_pano_y, raw_index, raw_obj_pred = output.full_res_panoptic_pred( super_index_level0_to_level1=nag[0].super_index, sub_level0_to_raw=nag[0].sub)

Also, is it posissible to not preprocess one point cloud into 4 seperate point clouds?

drprojects commented 3 months ago

how do i print or save information in format like [label, instance, [x,y,z ]] for all instances predicted

We provide load and save utilities for our data format, but not for the format you require. As you can imagine, I cannot provide support for any imaginable data format. So you will have to implement this yourself. Given the provided notebooks for full-resolution inference, this should be fairly straightforward.

also, panoptic predictions for kitti350 only show object and stuff and not label for that object. I am facing difficulty understanding that

I am not 100% sure what you mean but if you are wondering how to get the semantic segmentation labels, again, the provided notebook should answer all these questions.

if task_widget.value == 'panoptic':
    raw_pano_y, raw_index, raw_obj_pred = output.full_res_panoptic_pred(
        super_index_level0_to_level1=nag[0].super_index,
        sub_level0_to_raw=nag[0].sub)

raw_pano_y: the predicted semantic label raw_index: predicted instance index raw_obj_pred: same information but stored in an InstanceData object

These answers were available in the docstring of the full_res_panoptic_pred() function for instance.

Also, is it posissible to not preprocess one point cloud into 4 seperate point clouds?

Look into xy_tiling and pc_tiling in the project. You can change those from the config files to suit your need.

Overall, the answers to your questions were available in the documentation and in the code. We made sure the whole project is fairly commented, I invite your to have a look at the /docs and at the code comments to get a deeper understanding of how things work.

meehirmhatrepy commented 3 months ago

Thankyou so much for explaining in detail. It really helps.