Chen Yang comments

Results 14 comments of


                                            Chen Yang

Why is there no T-net in PointNet++ Architecture?

the stn3d and stnkd are T-nets

Semantic Segmentation Inference Take Too Long

Pointnet is indeed slow

Trained using the s3dis dataset, the calculated AP values for each region are 0

I have encountered the same situation as well, have you resolved it?

Trained using the s3dis dataset, the calculated AP values for each region are 0

ok, thx ---- Replied Message ---- From ***@***.***> Date 01/16/2025 00:27 To ***@***.***> Cc Chen ***@***.***>***@***.***> Subject Re: [JonasSchult/Mask3D] Trained using the s3dis dataset, the calculated AP values for each...

Inquiry on the Benefits of Gaussian Grouping Compared to SAM-based Object Reconstruction

> I have carefully studied your work on Gaussian Grouping and find it to be outstanding. However, I came across a similar approach that also leverages SAM technology to segment...

Training for variable number of points.

You should sample to a fixed number.

how to use the pre-trained part-seg model on my own pointcloud

You should first manually annotate your dataset.

Accuracy is low on the testing set (split from the same distributation of the training set) but is high when conducting K-Fold validation on the training set.

Can KFoldDataset be used for the dataset of the following architecture? data_prefix/ ├── class_x │ ├── xxx.png │ ├── xxy.png │ └── ... │ └── xxz.png └── class_y ├── 123.png...

Most of the indicators are 0 during training, and all of the indicators are -1 during verification.

the same question

python ./tools/pruning/get_channel_units.py /home/hy/mmpretrain/configs/mobilenet_v2/mobilenet-v2_8xb32_in1k.py -c -i --output-path ./temp

the same error