ModelNet40 中加入自己的数据集
ModelNet40 ：http://modelnet.cs.princeton.edu/#
含有40个内别的CAD三维模型，是评价点云深度学习模型进行语意分割、实例分割和分类的标准数据集
代码功能：对自己的点云目标进行标准化融合到ModleNet40中
主要步骤：
- 中心化：将点云中心移动到坐标原点；
- 尺度缩放：将所有点的坐标的绝对值限制在1以内；
- 采样：将点云采样到固定大小点数；
- 输出保存文件的格式；
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Wed Sep 25 21:53:03 2019
@author: sgl
"""
import os
import sys
import numpy as np
import h5py
BASE_DIR = os.path.dirname(os.path.abspath(__file__))
sys.path.append(BASE_DIR)
def getDataFiles(list_filename):
return [line.rstrip() for line in open(list_filename)]
def loadDataFile(path):
data = np.loadtxt(path)
num = data.shape[0]
point_xyz = data[:,0:3]
point_normal = data[:,3:6]
point_rgb = data[:,6:9]
# label just a example, should be repalced the real.
# modlenet40 is 0-39, so the label can be 40 and 41
label = np.ones((num,1), dtype=int)+39
return point_xyz, label
def change_scale(data):
#centre
xyz_min = np.min(data[:,0:3],axis=0)
xyz_max = np.max(data[:,0:3],axis=0)
xyz_move = xyz_min+(xyz_max-xyz_min)/2
data[:,0:3] = data[:,0:3]-xyz_move
#scale
scale = np.max(data[:,0:3])
data[:,0:3] = data[:,0:3]/scale
return data
def sample_data(data, num_sample):
""" data is in N x ...
we want to keep num_samplexC of them.
if N > num_sample, we will randomly keep num_sample of them.
if N < num_sample, we will randomly duplicate samples.
"""
N = data.shape[0]
if (N == num_sample):
return data, range(N)
elif (N > num_sample):
sample = np.random.choice(N, num_sample)
return data[sample, ...], sample
else:
sample = np.random.choice(N, num_sample-N)
dup_data = data[sample, ...]
return np.concatenate([data, dup_data], 0), list(range(N))+list(sample)
if __name__ == "__main__":
DATA_FILES =getDataFiles(os.path.join(BASE_DIR, 'file_path.txt'))
num_sample = 4096
DATA_ALL = []
for fn in range(len(DATA_FILES)):
current_data, current_label = loadDataFile(DATA_FILES[fn])
change_data = change_scale(current_data)
data_sample,index = sample_data(change_data, num_sample)
data_label = np.hstack((data_sample,current_label[index]))
DATA_ALL.append(data_label)
output = np.vstack(DATA_ALL)
output = output.reshape(-1,num_sample,4)
# train and test number, save data
if not os.path.exists('plant_train.h5'):
with h5py.File('plant_train.h5') as f:
f['data'] = output[0:7,0:3]
f['labels'] = output[0:8,4]
if not os.path.exists('plant_test.h5'):
with h5py.File('plant_test.h5') as f:
f['data'] = output[7:9,0:3]
f['labels'] = output[7:9,4]
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Wed Sep 25 16:42:25 2019
@author: sgl
"""
import os
import sys
import numpy as np
import h5py
BASE_DIR = os.path.dirname(os.path.abspath(__file__))
sys.path.append(BASE_DIR)
def sample_data(data, num_sample):
""" data is in N x ...
we want to keep num_samplexC of them.
if N > num_sample, we will randomly keep num_sample of them.
if N < num_sample, we will randomly duplicate samples.
"""
N = data.shape[0]
if (N == num_sample):
# return data, range(N)
return data
elif (N > num_sample):
sample = np.random.choice(N, num_sample)
# return data[sample, ...], sample
return data[sample, ...]
else:
sample = np.random.choice(N, num_sample-N)
dup_data = data[sample, ...]
# return np.concatenate([data, dup_data], 0), list(range(N))+list(sample)
return np.concatenate([data, dup_data], 0)
def creat_pcd_rgba(data,path):
#write pcd file
# path = os.path.join(BASE_DIR, 'out_put_'+str(1)+ '.pcd')
if os.path.exists(path):
os.remove(path)
Output_Data = open(path, 'a')
# headers
Output_Data.write('# .PCD v0.7 - Point Cloud Data file format\nVERSION 0.7\nFIELDS x y z rgba\nSIZE 4 4 4 4\nTYPE F F F U\nCOUNT 1 1 1 1')
string = '\nWIDTH ' + str(data.shape[0])
Output_Data.write(string)
Output_Data.write('\nHEIGHT 1\nVIEWPOINT 0 0 0 1 0 0 0')
string = '\nPOINTS ' + str(data.shape[0])
Output_Data.write(string)
Output_Data.write('\nDATA ascii')
# pack RGB
for j in range(data.shape[0]):
string = ('\n' + str(data[j,0]) + ' ' + str(data[j,1]) + ' ' +str(data[j,2]) + ' ' + str(int(data[j,3])))
Output_Data.write(string)
Output_Data.close()
def creat_pcd_rgb(data,path):
if os.path.exists(path):
os.remove(path)
Output_Data = open(path, 'a')
# headers
Output_Data.write('# .PCD v0.7 - Point Cloud Data file format\nVERSION 0.7\nFIELDS x y z rgba\nSIZE 4 4 4 4\nTYPE F F F U\nCOUNT 1 1 1 1')
string = '\nWIDTH ' + str(data.shape[0])
Output_Data.write(string)
Output_Data.write('\nHEIGHT 1\nVIEWPOINT 0 0 0 1 0 0 0')
string = '\nPOINTS ' + str(data.shape[0])
Output_Data.write(string)
Output_Data.write('\nDATA ascii')
# pack RGB
for j in range(data.shape[0]):
R=data[j,3]
G=data[j,4]
B=data[j,5]
value = (int(R) << 16 | int(G) << 8 | int(B))
string = ('\n' + str(data[j,0]) + ' ' + str(data[j,1]) + ' ' +str(data[j,2]) + ' ' + str(value))
Output_Data.write(string)
Output_Data.close()
def creat_txt(data,path):
np.savetxt(path,data)
data = np.loadtxt("/media/sgl/数据(混合硬盘)/pcl_ubuntu/pcl_test/plant.txt")
#centre
xyz_min = np.min(data[:,0:3],axis=0)
xyz_max = np.max(data[:,0:3],axis=0)
xyz_move = xyz_min+(xyz_max-xyz_min)/2
data[:,0:3] = data[:,0:3]-xyz_move
#scale
scale = np.max(data[:,0:3])
data[:,0:3] = data[:,0:3]/scale
#sample
data = sample_data(data,4096)
creat_pcd_rgba(data,'/media/sgl/数据(混合硬盘)/PointNet/basisnet/out_put_1.pcd')
creat_txt(data,'/media/sgl/数据(混合硬盘)/PointNet/basisnet/out_put_1.txt')
