implementation of PointNet using Paddle(classification only).
sklearn>=0.24.0
paddle>=0.2.0
tqdm
Total params: 3,483,377
Trainable params: 3,459,569
Non-trainable params: 23,808
use AdamW as optimizer with learn rate 0.0001
Model | Accuracy |
---|---|
PointNet(Offical) | 89.2 |
PointNet(PyTorch) | 90.6 |
PointNet(Paddle) | 91.2 |
Download datasets here
train : 9843
test : 2468
-
download dataset and unzip it into 'data'.
-
run 'train.py'
-
check points will be saved in 'check_points'