pointnet2InferenceOwnModel - 使用pointnet2推理自己的模型(以obj模型数据为例)

0 推理结果

1 运行pointnet2

https://github.com/yanx27/Pointnet_Pointnet2_pytorch#part-segmentation-shapenet

下载s3dis数据：
https://docs.google.com/forms/d/e/1FAIpQLScDimvNMCGhy_rmBA2gHfDu3naktRm6A8BPwAWWDv-Uhm6Shw/viewform?c=0&w=1

2 准备自己的测试数据

思路：将自己的一个obj模型，包装成为s3dis的一样的格式

方式如下：

# 1 准备原始数据：
目录结构如下：smallRoom.txt的每一行为 x y z  r g b, floor_1.txt 是 smallRoom.txt的复制
nash5@gas:~/prjs/Pointnet_Pointnet2_pytorch/data/s3dis$ tree ./Stanford3dDataset_v1.2_Aligned_Version
./Stanford3dDataset_v1.2_Aligned_Version
└── Area_5
    └── office_1
        ├── Annotations
        │   └── floor_1.txt
        └── smallRoom.txt

# 2 使用s3dis脚本生成测试数据：（你需要修改对应的meta文件，只留下一行）
nash5@gas:~/prjs/Pointnet_Pointnet2_pytorch/data/s3dis$ cat ../../data_utils/meta/anno_paths.txt 
Area_5/office_1/Annotations
nash5@gas:~/prjs/Pointnet_Pointnet2_pytorch/data/s3dis$ 

然后：
cd data_utils
python collect_indoor3d_data.py

之后会在data目录下生产stanford_indoor3d目录，将其移动到
3dis下面：
nash5@gas:~/prjs/Pointnet_Pointnet2_pytorch/data/s3dis$ ls ..
modelnet40_normal_resampled  s3dis      shapenetcore_partanno_segmentation_benchmark_v0_normal  stanford_indoor3d_ori
myData                       s3dis_ori  
nash5@gas:~/prjs/Pointnet_Pointnet2_pytorch/data/s3dis$ ls
Stanford3dDataset_v1.2_Aligned_Version  stanford_indoor3d

# 3 运行测试脚本：（你需要改动batch_size，1080可以设置为16），然后你会在log里的几层嵌套中找到这个
nash5@gas:~/prjs/Pointnet_Pointnet2_pytorch$ python test_semseg.py --log_dir pointnet2_sem_seg --test_area 5 --visual --batch_size 16
ay(total_correct_class_tmp) / (np.array(total_iou_deno_class_tmp, dtype=np.float) + 1e-6)
[0.         0.03883974 0.         0.         0.         0.
 0.         0.         0.         0.         0.         0.
 0.        ]
Mean IoU of Area_5_office_1: 0.0388
----------------------------
test_semseg.py:185: DeprecationWarning: `np.float` is a deprecated alias for the builtin `float`. To silence this warning, use `float` by itself. Doing this will not modify any behavior and is safe. If you specifically wanted the numpy scalar type, use `np.float64` here.
Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations
  IoU = np.array(total_correct_class) / (np.array(total_iou_deno_class, dtype=np.float) + 1e-6)
test_semseg.py:190: RuntimeWarning: invalid value encountered in true_divide
  total_correct_class[l] / float(total_iou_deno_class[l]))
------- IoU --------
class ceiling       , IoU: 0.000 
class floor         , IoU: 0.039 
class wall          , IoU: 0.000 
class beam          , IoU: 0.000 
class column        , IoU: 0.000 
class window        , IoU: 0.000 
class door          , IoU: 0.000 
class table         , IoU: nan 
class chair         , IoU: 0.000 
class sofa          , IoU: 0.000 
class bookcase      , IoU: 0.000 
class board         , IoU: 0.000 
class clutter       , IoU: 0.000 

eval point avg class IoU: 0.002988
test_semseg.py:194: DeprecationWarning: `np.float` is a deprecated alias for the builtin `float`. To silence this warning, use `float` by itself. Doing this will not modify any behavior and is safe. If you specifically wanted the numpy scalar type, use `np.float64` here.
Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations
  np.mean(np.array(total_correct_class) / (np.array(total_seen_class, dtype=np.float) + 1e-6))))
eval whole scene point avg class acc: 0.002988
eval whole scene petails and guidance: https://numpy.org/devdocs/release

# 4 查看结果： 用meshlab看那个pred.obj
nash5@gas:~/prjs/Pointnet_Pointnet2_pytorch$ ls log/sem_seg/pointnet2_sem_seg/visual/Area_5_office_1
Area_5_office_1_gt.obj    Area_5_office_1_pred.obj  Area_5_office_1.txt