pointcloud_homework1

部分代码参考自https://blog.youkuaiyun.com/program_developer/article/details/80632779与https://colab.research.google.com/drive/17uvogSe7dufrjpslM3X0yvcfEc6e-9iv

task

1. Build dataset
   a. Download ModelNet40 dataset
   b. Select one point cloud from each category
2. Perform PCA for the 40 objects, visualize it.
3. Perform surface normal estimation for each point of each object, visualize it.
4. Downsample each object using voxel grid downsampling (exact, both centroid &
   random). Visualize the results.

solutions

使用google colab来解答，使用pycharm来进行部分的可视化
在谷歌的云端硬盘新建一个google colabortary进行如下操作，并在pycharm新建一个用于可视化的py文件

---

1.a
下载数据集

把数据集下载到与该ipynb文件同一位置的data文件夹内

import os
import sys
import numpy as np
import h5py
BASE_DIR = os.path.dirname(os.path.abspath(__file__))
sys.path.append(BASE_DIR)
# Download dataset for point cloud classification
DATA_DIR = os.path.join(BASE_DIR, 'data')
if not os.path.exists(DATA_DIR):    
	os.mkdir(DATA_DIR)
if not os.path.exists(os.path.join(DATA_DIR, 'modelnet40_ply_hdf5_2048')):
 	www = 'https://shapenet.cs.stanford.edu/media/modelnet40_ply_hdf5_2048.zip'
 	zipfile = os.path.basename(www)    
 	os.system('wget %s; unzip %s' % (www, zipfile))  
 	os.system('mv %s %s' % (zipfile[:-4], DATA_DIR)) 
 	os.system('rm %s' % (zipfile))

1.b
读取数据

提供相关读取函数

import os
import sys
import numpy as np
import h5py

def shuffle_data(data, labels):
    """ Shuffle data and labels.
        Input:
          data: B,N,... numpy array
          label: B,... numpy array
        Return:
          shuffled data, label and shuffle indices
    """
    idx = np.arange(len(labels))
    np.random.shuffle(idx)
    return data[idx, ...], labels[idx], idx


def rotate_point_cloud(batch_data):
    """ Randomly rotate the point clouds to augument the dataset
        rotation is per shape based along up direction
        Input:
          BxNx3 array, original batch of point clouds
        Return:
          BxNx3 array, rotated batch of point clouds
    """
    rotated_data = np.zeros(batch_data.shape, dtype=np.float32)
    for k in range(batch_data.shape[0]):
        rotation_angle = np.random.uniform() * 2 * np.pi
        cosval = np.cos(rotation_angle)
        sinval = np.sin(rotation_angle)
        rotation_matrix = np.array([[cosval, 0, sinval],
                                    [0, 1, 0],
                                    [-sinval, 0, cosval]])
        shape_pc = batch_data[k, ...]
        rotated_data[k, ...] = np.dot(shape_pc.reshape((-1, 3)), rotation_matrix)
    return rotated_data


def rotate_point_cloud_by_angle(batch_data, rotation_angle):
    """ Rotate the point cloud along up direction with certain angle.
        Input:
          BxNx3 array, original batch of point clouds
        Return:
          BxNx3 array, rotated batch of point clouds
    """
    rotated_data = np.zeros(batch_data.shape, dtype=np.float32)
    for k in range(batch_data.shape[0]):
        #rotation_angle = np.random.uniform() * 2 * np.pi
        cosval = np.cos(rotation_angle)
        sinval = np.sin(rotation_angle)
        rotation_matrix = np.array([[cosval, 0, sinval],
                                    [0, 1, 0],
                                    [-sinval, 0, cosval]])
        shape_pc = batch_data[k, ...]
        rotated_data[k, ...] = np.dot(shape_pc.reshape((-1, 3)), rotation_matrix)
    return rotated_data


def jitter_point_cloud(batch_data, sigma=0.01, clip=0.05):
    """ Randomly jitter points. jittering is per point.
        Input:
          BxNx3 array, original batch of point clouds
        Return:
          BxNx3 array, jittered batch of point clouds
    """
    B, N, C = batch_data.shape
    assert(clip > 0)
    jittered_data = np.clip(sigma * np.random.randn(B, N, C), -1*clip, clip)
    jittered_data += batch_data
    return jittered_data

def getDataFiles(list_filename):
    return [line.rstrip() for line in open(list_filename)]

def load_h5(h5_filename):
    f = h5py.File(h5_filename)
    data = f['data'][:]
    label = f['label'][:]
    return (data, label)

def loadDataFile(filename):
    return load_h5(filename)

def load_h5_data_label_seg(h5_filename):
    f = h5py.File(h5_filename)
    data = f['data'][:]
    label = f['label'][:]
    seg = f['pid'][:]
    return (data, label, seg)


def loadDataFile_with_seg(filename):
    return load_h5_data_label_seg(filename)

读取数据，代码只读取了指定0~39编号中的0号类别的一个数据

import argparse
import math
import h5py
import numpy as np
import tensorflow as tf
import socket
import importlib
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
import os
import sys
BASE_DIR = os.path.dirname(os.path.realpath('__file__'))
sys.path.append(BASE_DIR)



# ModelNet40 official train/test split
TRAIN_FILES = getDataFiles( \
    os.path.join(BASE_DIR, 'data/modelnet40_ply_hdf5_2048/train_files.txt'))
TEST_FILES = getDataFiles(\
    os.path.join(BASE_DIR, 'data/modelnet40_ply_hdf5_2048/test_files.txt'))
CLASS_NAME=getDataFiles(\
    os.path.join(BASE_DIR, 'data/modelnet40_ply_hdf5_2048/shape_names.txt'))


current_data, current_label = loadDataFile(os.path.join(BASE_DIR,TRAIN_FILES[0]))

def points_to_pcd(points,filename):
    xlist=[p[0] for p in points]
    ylist=[p[1] for p in points]
    zlist=[p[2] for p in points]
    if not os.path.exists(filename):
        f = open(filename,'w')
        f.close()
    with open(filename, 'w') as file_to_write:
        file_to_write.writelines("# .PCD v0.7 - Point Cloud Data file format\n")
        file_to_write.writelines("VERSION 0.7\n")
        file_to_write.writelines("FIELDS x y z\n")
        file_to_write.writelines("SIZE 4 4 4\n")
        file_to_write.writelines("TYPE F F F\n")
        file_to_write.writelines("COUNT 1 1 1\n")
        file_to_write.writelines("WIDTH " + str(len(xlist)) + "\n")
        file_to_write.writelines("HEIGHT 1\n")
        file_to_write.writelines("VIEWPOINT 0 0 0 1 0 0 0\n")
        file_to_write.writelines("POINTS " + str(len(xlist)) + "\n")
        file_to_write.writelines("DATA ascii\n")
        for i in range(len(xlist)):
            file_to_write.writelines(str(xlist[i]) + " " + str(ylist[i]) + " " + str(zlist[i]) + "\n")

cname=0
id=list(current_label[:,0]).index(cname)
#将点云数据存储，并使用pycharm中open3d可视化
points_to_pcd(current_data[id],'/content/drive/My Drive/orign.pcd')

import open3d as o3d
#path为导出的pcd本地路径
pcd = o3d.io.read_point_cloud(path)
o3d.visualization.draw_geometries([pcd])

2.
PCA降维

def pca(X, k):  # k is the components you want
    # mean of each feature
    #X为一个m×n的矩阵，n为原维度
    n_samples, n_features = X.shape
    mean = np.array([np.mean(X[:, i]) for i in range(n_features)])
    # normalization
    norm_X = X - mean
    # scatter matrix
    scatter_matrix = np.dot(np.transpose(norm_X), norm_X)/X.shape[0]
    # Calculate the eigenvectors and eigenvalues
    eig_val, eig_vec = np.linalg.eig(scatter_matrix)
    eig_pairs = [(np.abs(eig_val[i]), eig_vec[:, i]) for i in range(n_features)]
    # sort eig_vec based on eig_val from highest to lowest
    eig_pairs.sort(reverse=True)
    # select the top k eig_vec
    feature = np.array([ele[1] for ele in eig_pairs[:k]])
    # get new data
    data = np.dot(norm_X, np.transpose(feature))
    return data

res=pca(current_data[id],2)

xset=res[:,0]
yset=res[:,1]

fig = plt.figure()
plt.scatter(xset, yset, c='k')
plt.axis('off')
plt.title(CLASS_NAME[current_label[id][0]])

plt.show()

3.
计算法向量并可视化

def normal_pca(X):  # k is the components you want
    # mean of each feature
    #X为一个m×n的矩阵，n为原维度
    n_samples, n_features = X.shape
    mean = np.array([np.mean(X[:, i]) for i in range(n_features)])
    # normalization
    norm_X = X - mean
    # scatter matrix
    scatter_matrix = np.dot(np.transpose(norm_X), norm_X)
    # Calculate the eigenvectors and eigenvalues
    eig_val, eig_vec = np.linalg.eig(scatter_matrix)
    eig_pairs = [(np.abs(eig_val[i]), eig_vec[:, i]) for i in range(n_features)]
    # sort eig_vec based on eig_val from highest to lowest
    eig_pairs.sort(reverse=False)
    # select the top k eig_vec
    feature = np.array([ele[1] for ele in eig_pairs[:1]])
    # get new data

    return feature[0]


def PointsWithNormal_to_pcd(points,mormallist,filename):
    xlist=[p[0] for p in points]
    ylist=[p[1] for p in points]
    zlist=[p[2] for p in points]
    #normallist=[p[3] for p in points]
    if not os.path.exists(filename):
        f = open(filename,'w')
        f.close()
    with open(filename, 'w') as file_to_write:
        file_to_write.writelines("# .PCD v0.7 - Point Cloud Data file format\n")
        file_to_write.writelines("VERSION 0.7\n")
        file_to_write.writelines("FIELDS x y z normal_x normal_y normal_z\n")
        file_to_write.writelines("SIZE 4 4 4 4 4 4\n")
        file_to_write.writelines("TYPE F F F F F F\n")
        file_to_write.writelines("COUNT 1 1 1 1 1 1\n")
        file_to_write.writelines("WIDTH " + str(len(xlist)) + "\n")
        file_to_write.writelines("HEIGHT 1\n")
        file_to_write.writelines("VIEWPOINT 0 0 0 1 0 0 0\n")
        file_to_write.writelines("POINTS " + str(len(xlist)) + "\n")
        file_to_write.writelines("DATA ascii\n")
        for i in range(len(xlist)):
            file_to_write.writelines(str(xlist[i]) + " " + str(ylist[i]) + " " + str(zlist[i]) + " " +str(mormallist[i][0]) + " " + str(mormallist[i][1]) + " " + str(mormallist[i][2]) + "\n")

def findNeighborsInRadius(pointcloud,point,radius):
    neighbors=[]
    for p in pointcloud:
        if np.linalg.norm(p-point)<radius:
            neighbors.append(p)
    return neighbors


temp=current_data[id]
normalset=[]
for k in temp:
    neighbourset=findNeighborsInRadius(temp,k,0.1)
    neighbourset=np.array(neighbourset)
    normal=normal_pca(neighbourset)
    normalset.append(list(normal))


PointsWithNormal_to_pcd(temp,normalset,"/content/drive/My Drive/orignwithnormal.pcd")

pycharm可视化

import open3d as o3d
#path为导出的pcd本地路径
pcd = o3d.io.read_point_cloud(path)
o3d.visualization.draw_geometries([pcd])

下采样

#随机去除点
import random

cname=0
id=list(current_label[:,0]).index(cname)


index = random.sample(range(0,2048),500)
origindata=current_data[id]
randomdata=current_data[id][index]

xset=origindata[:,0]
yset=origindata[:,1]
zset=origindata[:,2]

fig = plt.figure()
ax = Axes3D(fig)
ax.scatter(xset, yset, zset,c='k')
plt.axis('off')
plt.title(CLASS_NAME[current_label[id][0]])

xset=randomdata[:,0]
yset=randomdata[:,1]
zset=randomdata[:,2]

fig = plt.figure()
ax = Axes3D(fig)
ax.scatter(xset, yset, zset,c='k')
plt.axis('off')
plt.title(CLASS_NAME[current_label[id][0]])


plt.show()

#voxelgrid去除中心点

def CentriodVoxelGridDownsample(pointcloud,size):
    # 计算每个维度上的极值
    x_min=min(pointcloud[:][0])
    x_max=max(pointcloud[:][0])
    y_min=min(pointcloud[:][1])
    y_max=max(pointcloud[:][1])
    z_min=min(pointcloud[:][2])
    z_max=max(pointcloud[:][2])
    Dx=int((x_max-x_min)/size)
    Dy=int((y_max-y_min)/size)
    Dz=int((z_max-z_min)/size)
    # 用字典存储h及其对应的点
    d={}
    CentriodPoints=[]
    for p in pointcloud:
        
        hx=int((p[0]-x_min)/size)
        hy=int((p[1]-y_min)/size)
        hz=int((p[2]-z_min)/size)
        h=(hx+hy*Dx+hz*Dx*Dy)%90
        
        if h not in d.keys():
            d[h]=[]
            d[h].append(hx)
            d[h].append(hy)
            d[h].append(hz)
            d[h].append(p)
        else:
            if hx!=d[h][0] and hy!=d[h][1] and hz!=d[h][2]:
                d[h]=d[h][3:]
                CentriodPoint=np.mean(d[h],axis=0)
                CentriodPoints.append(CentriodPoint)
                d[h]=[]
                d[h].append(hx)
                d[h].append(hy)
                d[h].append(hz)
                d[h].append(p)
            else:
                d[h].append(p)

    # 求每个h对应的点的中心
    
    
    for key,value in d.items():
        value=value[3:]
        
        CentriodPoint=np.mean(value,axis=0)
        CentriodPoints.append(CentriodPoint)

    CentriodPoints=np.array(CentriodPoints)
    return CentriodPoints


cname=0
id=list(current_label[:,0]).index(cname)

rescen=CentriodVoxelGridDownsample(current_data[id],0.05)


xset=rescen[:,0]
yset=rescen[:,1]
zset=rescen[:,2]

fig = plt.figure()
ax = Axes3D(fig)
ax.scatter(xset, yset, zset,c='k')
plt.axis('off')
plt.title(CLASS_NAME[current_label[id][0]])


plt.show()

#voxelgrid随机去点
def RandomVoxelGridDownsample(pointcloud,size):
    # 计算每个维度上的极值
    x_min=min(pointcloud[:][0])
    x_max=max(pointcloud[:][0])
    y_min=min(pointcloud[:][1])
    y_max=max(pointcloud[:][1])
    z_min=min(pointcloud[:][2])
    z_max=max(pointcloud[:][2])
    Dx=int((x_max-x_min)/size)
    Dy=int((y_max-y_min)/size)
    Dz=int((z_max-z_min)/size)
    # 用字典存储h及其对应的点
    d={}
    RandomPoints=[]
    for p in pointcloud:
        hx=int((p[0]-x_min)/size)
        hy=int((p[1]-y_min)/size)
        hz=int((p[2]-z_min)/size)
        h=(hx+hy*Dx+hz*Dx*Dy)%90

        if h not in d.keys():
            d[h]=[]
            d[h].append(hx)
            d[h].append(hy)
            d[h].append(hz)
            d[h].append(p)
        else:
            if hx!=d[h][0] and hy!=d[h][1] and hz!=d[h][2]:
                d[h]=d[h][3:]
                RandomPoint=random.choice(d[h])
                RandomPoints.append(RandomPoint)
                d[h]=[]
                d[h].append(hx)
                d[h].append(hy)
                d[h].append(hz)
                d[h].append(p)
            else:
                d[h].append(p)

        
    # 求每个h对应的点的中心
    
    for key,value in d.items():
        value=value[3:]
        RandomPoint=random.choice(value)
        RandomPoints.append(RandomPoint)

    RandomPoints=np.array(RandomPoints)
    return RandomPoints


cname=0
id=list(current_label[:,0]).index(cname)

rescen=CentriodVoxelGridDownsample(current_data[id],0.05)


xset=rescen[:,0]
yset=rescen[:,1]
zset=rescen[:,2]

fig = plt.figure()
ax = Axes3D(fig)
ax.scatter(xset, yset, zset,c='k')
plt.axis('off')
plt.title(CLASS_NAME[current_label[id][0]])


plt.show()