Xiaoyang Wu comments

Results 495 comments of


                                            Xiaoyang Wu

How is the receptive field in Figure 13 of OA-CNNs visualized?

@Pbihao

Train PPT but got Loss = nan

Try again or set amp False. It does occur from time to time. Maybe clip loss within a range can solve it but I have not try it.

Visualize nuscenes train results

You have to write a script to map these predictions into a color and then export them to a ply file. There is some tool in pointcept.utils.visualization.save_point_cloud.

Visualize nuscenes train results

I think the order of points is the same as processed data. You can get the coordinate we dataset by the data name?

Visualize nuscenes train results

Hey all, hope our recently demo code for outdoor point cloud is helpful: https://github.com/Pointcept/Concerto/blob/main/demo/0_pca_outdoor.py

OA-CNNs training S3DIS and KITTI, as well as nuScenes dataset

Hi, getting configs for each other dataset is pretty easy. Just copy our config for SpUNet for each dataset and them replace the model parameters with our provided.

OA-CNNs training S3DIS and KITTI, as well as nuScenes dataset

Hi, 0.05 for outdoor. Use config for SpUNet as a base, I remember I also add the Lovasz loss.

OA-CNNs training S3DIS and KITTI, as well as nuScenes dataset

I remember that it should be the same. (I didn’t have the memory that changed the parameter for the outdoor case)

OA-CNNs training S3DIS and KITTI, as well as nuScenes dataset

Let me introduce how to custom such a config (take semantic-kitti as an example): 1. Copy one config like 1semseg-spunet-v1m1-base1 and rename it as 1semseg-oacnn-v1m1-base1. 2. Compare the config difference...

Memory issue during testing phase with custom dataset for semantic segmentation

Left one for inference, don't make it empty.