项目文件夹

文件

T

文件历史

esang 3fef5d27d3 [Model] PCT (#3339 )

* publish pct

* add train_cls

* add readme

* update opt for point transformer

* update the example index

* update for comments

Co-authored-by: Tong He <hetong007@gmail.com>

2021-09-13 17:17:24 +08:00

helper.py

[Model] Point transformer (#3284 )

2021-08-27 14:33:18 +08:00

ModelNetDataLoader.py

[Model] Point transformer (#3284 )

2021-08-27 14:33:18 +08:00

point_transformer.py

[Model] Point transformer (#3284 )

2021-08-27 14:33:18 +08:00

provider.py

[Model] Point transformer (#3284 )

2021-08-27 14:33:18 +08:00

README.md

[Model] Point transformer (#3284 )

2021-08-27 14:33:18 +08:00

ShapeNet.py

[Model] Point transformer (#3284 )

2021-08-27 14:33:18 +08:00

train_cls.py

[Model] PCT (#3339 )

2021-09-13 17:17:24 +08:00

train_partseg.py

[Model] PCT (#3339 )

2021-09-13 17:17:24 +08:00

README.md

Point Transformer

This model is implemented on August 27, 2021 when there is no official code released.
Thus we implemented this model based on the code from https://github.com/qq456cvb/Point-Transformers.

This is a reproduction of the paper: Point Transformer.

Performance

Task	Dataset	Metric	Score - Paper	Score - DGL (Adam)	Score - DGL (SGD)	Time(s) - DGL
Classification	ModelNet40	Accuracy	93.7	92.0	91.5	117.0
Part Segmentation	ShapeNet	mIoU	86.6	84.3	85.1	260.0

Time(s) are the average training time per epoch, measured on EC2 p3.8xlarge instance w/ Tesla V100 GPU.

How to Run

For point cloud classification, run with

python train_cls.py --opt [sgd/adam]

For point cloud part-segmentation, run with

python train_partseg.py --opt [sgd/adam]