search

EnyaHermite / SPH3D-GCN

Spherical Kernel for Efficient Graph Convolution on 3D Point Clouds
MIT License
169 stars 32 forks source link

training and inference #10

Open abhigoku10 opened 4 years ago

abhigoku10 commented 4 years ago

@hlei-ziyan thanks for open sourcing the wonderfull work , i had few queries Q1 have you trained the architecture on the available other dataset like semanttic Kitti and 3D dataset Q2 If not trained can we follow the same training pipeline , if trained can you please share the pre-trained model Q3 can we use the currently pre-trained model to test on custom dataset which less number of point cloud density

Thanks in advance

EnyaHermite commented 4 years ago

A1: Not yet. A2: Yes, the new tensorflow modules we introduce is applicable to any point cloud dataset. You just need to prepare your data (block pre-splitting or random block sampling on the fly is the current convention). The training/evaluation will be rather similar to those of S3DIS and ScanNet. A3: About applying pre-trained models to test with less or more number of point cloud density, short answer is that testing pre-trained model on more points is fine, but the performance drops quickly for less points. Our cvpr2020 paper (should be available after this March) studied this detailedly. If you are interested in using trained-model across different number of points, the fuzzy kernel proposed in the cvpr2020 paper suits your purpose (code now partially available in my github repository 'SegGCN').