RandLA-Net复现S3DIS记录

S3DIS

环境安装：[https://blog.csdn.net/Parkerfive/article/details/123431266]
参考：在RTX 3090上复现点云语义分割算法RandLA-net
Ubuntu18.04/20.04复现算法RandLa-net 数据集S3DIS

1. 运行 python utils/data_prepare_s3dis.py

报错1：
FileNotFoundError: [Errno 2] No such file or directory: ‘/data/S3DIS/original_ply’
找到data_prepare_s3dis.py，将/data/S3DIS/Stanford3dDataset_v1.2_Aligned_Version
改成绝对路径’/home/clm/training/RandLA-Net-master/data/S3DIS/Stanford3dDataset_v1.2_Aligned_Version ’

报错2：
module ‘tensorflow’ has no attribute ‘py_func’
参考：https://www.codenong.com/cs106162826/
把tf.py_func改成tf.compat.v1.py_func

报错3：
Traceback (most recent call last):
File “main_S3DIS.py”, line 235, in
dataset.init_input_pipeline()
File “main_S3DIS.py”, line 214, in init_input_pipeline
iter = tf.data.Iterator.from_structure(self.batch_train_data.output_types, self.batch_train_data.output_shapes)
AttributeError: type object ‘IteratorBase’ has no attribute ‘from_structure’
keras版本不兼容

报错4：
AttributeError: ‘NoneType’ object has no attribute ‘name’
main_S3DIS.py修改路径为绝对路径

就这个路径，修改了好几次，一样的东西却没有跑通，今天不知道怎么的突然就跑通了

运行 python -B main_S3DIS.py --gpu 0 --mode train --test_area 1

成功加载数据了，也没有报错了

epoch0这里一直显示不动，过了很久才跑起来了，还以为是哪里出错了呢，丢在一旁突然就跑通了。

运行 sh jobs_6_fold_cv_s3dis.sh
最后的结果
EPOCH 99

Step 00049550 L_out=0.840 Acc=0.96 ---  379.55 ms/batch
Step 00049600 L_out=0.942 Acc=0.96 ---  449.07 ms/batch
Step 00049650 L_out=1.053 Acc=0.95 ---  401.10 ms/batch
Step 00049700 L_out=1.087 Acc=0.94 ---  427.35 ms/batch
Step 00049750 L_out=0.780 Acc=0.96 ---  390.87 ms/batch
Step 00049800 L_out=0.741 Acc=0.96 ---  381.06 ms/batch
Step 00049850 L_out=0.698 Acc=0.97 ---  424.08 ms/batch
Step 00049900 L_out=0.642 Acc=0.96 ---  474.82 ms/batch
Step 00049950 L_out=1.196 Acc=0.95 ---  382.93 ms/batch
Step 00050000 L_out=0.710 Acc=0.97 ---  340.56 ms/batch
eval accuracy: 0.8745440795898437
mean IOU:0.7105611018660913
Mean IoU = 71.1%
--------------------------------------------------------------------------------------
71.06 | 96.33 94.63 77.06 53.56 50.45 75.24 82.60 65.57 77.70 58.94 60.65 64.69 66.31 
--------------------------------------------------------------------------------------

Best m_IoU is: 71.366