Skip to content

Latest commit

 

History

History
65 lines (48 loc) · 2.83 KB

README.md

File metadata and controls

65 lines (48 loc) · 2.83 KB

PointNetProject

Final project of Geometric Learning Course - Technion 048865

Introducion

This project was implemented by PyTorch. Models are traind on ModelNet40 dataset. A parser is use for the train and eval input (insturction below). Some new lifting enhancmented are suggested as part of this project, including: use of point normals, use 3 order moments and use of points curvature. Two basic models are available:

  1. PointNet cls
  2. Momenet cls

Use Train.py to train the desired model. Use evaluate.py to test a trained model.

Train Running Instructions

Available models:

  1. pointnet_cls - Basic PointNet classification network [1]
  2. momenet_cls - Basic Momenet classification network [2]
  3. momenet_with_normals - Momenet classification network using normals as well.

Train Input Params

Parameter Name Description Possible values
model used model type pointnet_cls \ momenet_cls \ momenet_with_normals
test_name save name for logs string
log_dir path to log dir string
num_point Num of point in point cloud [256/512/1024/2048] [default: 1024]
max_epoch Number of epochs int
batch_size train batch size int [default: 32]
use_knn Use Knn 0- False 1 - True
moment_order Maximal moment order 1 / 2 / 3
use_lifting Use curvature lifting 0- False 1 - True

Train Usage Example

python train.py --max_epoch 100 --model "momenet_with_normals" --use_knn 1 --test_name "TstName" --use_lifting 1 --moment_order 3

Evaluate Running Instructions

You can choose one of the pretrained models saved as .pth files.

Evaluate Input Params

Parameter Name Description Possible values
model used model type pointnet_cls \ momenet_cls \ momenet_with_normals
test_name save name for logs string
log_dir path to log dir string
num_point Num of point in point cloud [256/512/1024/2048] [default: 1024]
use_knn Use Knn 0- False 1 - True
moment_order Maximal moment order 1 / 2 / 3
use_lifting Use curvature lifting 0- False 1 - True
model_path Path to a .pth file string

Evaluate Usage Example

python train.py --model_path "weights/momenet_3Order.pth" --model "momenet_cls"

List Of Saved Modles

All saved models can be found in log dir

Refrences

[1] Qi, Charles R and Su, Hao and Mo, Kaichun and Guibas, Leonidas, "Pointnet: Deep learning on point sets for 3d classification and segmentation".Proceedings of the IEEE conference on computer vision and pattern recognition, p652--660,2017.

[2] Joseph-Rivlin, Mor and Zvirin, Alon and Kimmel, Ron, "Momen (e) t: Flavor the moments in learning to classify shapes",Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops,2019