三维深度学习（二）win10 PointNet pytorch版本测试

1. 环境

win10

Anaconda3.6

Pytorch1.4

2. PointNet pytorch下载，并下载好数据集

https://github.com/fxia22/pointnet.pytorch

3. Open Terminal

打开Anaconda Navigator——ENvironmens——Pytorch_envs(自己创建的虚拟环境）

左键点击三角图标-Open Terminal

4. 运行命令

(1）进入到pointnet代码文件夹主目录，输入以下命令，安装环境

pip install -e .

(2) 进入utils文件夹

训练 train_classification.py

python train_classification.py --dataset=D:\work\paper\0323PointNet\pointnet.pytorch-master\shapenetcore_partanno_segmentation_benchmark_v0\ --nepoch=5 --dataset_type=shapenet

备注：

报错1：Detected call of `lr_scheduler.step()` before `optimizer.step()`

解决办法：https://blog.csdn.net/zfjBIT/article/details/105114526

报错2：The "freeze_support()" line can be omitted if the program is not going to be frozen to produce an executable.

解决办法：https://blog.csdn.net/zfjBIT/article/details/105117206

报错3：PermissionError: [WinError 5] 拒绝访问

解决办法：https://blog.csdn.net/zfjBIT/article/details/105121116

训练结果：

 

会生成cls文件夹以及model 

训练 train_segmentation.py 

python train_segmentation.py --dataset=D:\work\paper\0323PointNet\pointnet.pytorch-master\shapenetcore_partanno_segmentation_benchmark_v0\ --nepoch=5

报错：

[0: 0/83] train loss: 1.375747 accuracy: 0.370837
[0: 0/83] [94mtest[0m loss: 1.389753 accuracy: 0.490475
Traceback (most recent call last):
  File "train_segmentation.py", line 88, in 
    pred, trans, trans_feat = classifier(points)
  File "E:\softwareInstallation\Anaconda3\envs\Pytorch_envs\lib\site-packages\torch\nn\modules\module.py", line 532, in __call__
    result = self.forward(*input, **kwargs)
  File "d:\work\paper\0323pointnet\pointnet.pytorch-master\pointnet\model.py", line 167, in forward
    x, trans, trans_feat = self.feat(x)
  File "E:\softwareInstallation\Anaconda3\envs\Pytorch_envs\lib\site-packages\torch\nn\modules\module.py", line 532, in __call__
    result = self.forward(*input, **kwargs)
  File "d:\work\paper\0323pointnet\pointnet.pytorch-master\pointnet\model.py", line 127, in forward
    return torch.cat([x, pointfeat], 1), trans, trans_feat
RuntimeError: CUDA out of memory. Tried to allocate 334.00 MiB (GPU 0; 6.00 GiB total capacity; 3.99 GiB already allocated; 320.14 MiB free; 4.29 GiB reserved in total by PyTorch)

将batchSize降低到8后可以运行（默认为32）：

python train_segmentation.py --dataset=D:\work\paper\0323PointNet\pointnet.pytorch-master\shapenetcore_partanno_segmentation_benchmark_v0\ --nepoch=5 --batchSize=8

训练结果： 

会生成seg文件夹以及model 

 

未完待续~~~~~~~~~~