Semantic-Kitti数据解析
1. .bin文件
文件中保存的是二进制格式的四维雷达数据，包括x,y,z,intensity
打开文件：
def load_data_points(points_path):
# cloud = np.fromfile(points_path, dtype=np.float32).reshape((-1, 4))
cloud = np.fromfile(points_path, dtype=np.float32)
return cloud2. .label文件
文件中保存对应每个点的label
打开文件：
def load_data_labels(label_path):
labels = np.fromfile(label_path, dtype=np.uint32).reshape((-1, 1))
return labels3. 我是切割自己数据集时发现坐标轴出现问题，训练的模型无法正确显示其效果，排查发现训练集数据xyz值做了归一化与原始数据不一致，导致应用模型时输入原始数据效果不佳。
重新加载所有训练集数据修正偏移量（我的是原始数据Z值减了1.34），代码如下：
import numpy as np
import argparse
import struct
import math
import os
EXTENSIONS_SCAN = ['.bin']
EXTENSIONS_LABEL = ['.label']
def is_scan(filename):
return any(filename.endswith(ext) for ext in EXTENSIONS_SCAN)
def is_label(filename):
return any(filename.endswith(ext) for ext in EXTENSIONS_LABEL)
def getitem(dataset_path, seq_id):
root = os.path.join(dataset_path, "sequences")
seq_id = '{0:02d}'.format(int(seq_id))
print(seq_id)
scan_files = []
label_files = []
for seq in seq_id:
scan_path = os.path.join(root, seq_id)
# print("scan_path:{}".format(scan_path))
label_path = os.path.join(root, seq_id)
scan_files = [os.path.join(dp, f) for dp, dn, fn in os.walk(
os.path.expanduser(scan_path)) for f in fn if is_scan(f)]
# print("scan_files:{}".format(scan_files))
label_files = [os.path.join(dp, f) for dp, dn, fn in os.walk(
os.path.expanduser(label_path)) for f in fn if is_label(f)]
scan_files.extend(scan_files)
label_files.extend(label_files)
return scan_files, label_files
def load_data_points(points_path):
# cloud = np.fromfile(points_path, dtype=np.float32).reshape((-1, 4))
cloud = np.fromfile(points_path, dtype=np.float32)
# print(cloud.shape)
return cloud
def load_data_labels(label_path):
labels = np.fromfile(label_path, dtype=np.uint32).reshape((-1, 1))
return labels
def transform(roll, pitch, yaw ,x, y, z):
Rx = np.array([[1, 0, 0],
[0, np.cos(roll), -(np.sin(roll))],
[0, np.sin(roll), np.cos(roll)]])
Ry = np.array([[np.cos(pitch), 0, np.sin(pitch)],
[0, 1, 0],
[-(np.sin(pitch)), 0 ,np.cos(pitch)]])
Rz = np.array([[np.cos(yaw), -(np.sin(yaw)), 0],
[np.sin(yaw), np.cos(yaw), 0],
[0, 0, 1]])
R = Rz.dot(Ry.dot(Rx))
t = np.array([x,y,z])
t = t.reshape(3,1)
Rt = np. vstack((np.hstack([R, t]), [0,0,0,1]))
print(Rt)
return Rt
def rechange_points(points):
points_num = int(points.size/4)
for num in range(points_num):
# print("----test_points_intensity:{}------".format(points[4*num+3]))
points[4*num+2] += 1.34
if(points[4*num+3] > 500):
points[4*num+3] = intensity[num]/1000
# print("----test_points_intensity:{}------".format(points[4*num+3]))
return points
def WriteFile(points, file_save):
binfile = open(file_save, 'wb') #二进制写模式
for wf in range(int(points.size)):
a = struct.pack('f',points[wf])
binfile.write(a)
binfile.close()
if __name__ == "__main__":
parser = argparse.ArgumentParser("./main.py")
parser.add_argument(
'--dataset_path', '-d',
type=str,
required=True,
)
parser.add_argument(
'--seq_id', '-s',
type=str,
required=True,
)
parser.add_argument(
'--dataset_path_save', '-ds',
type=str,
required=True,
)
parser.add_argument(
'--seq_id_save', '-ss',
type=str,
required=True,
)
FLAGS, unparsed = parser.parse_known_args()
print(FLAGS.dataset_path_save)
scan_files, label_files = getitem(FLAGS.dataset_path, FLAGS.seq_id)
scan_files_save, label_files_save = getitem(FLAGS.dataset_path_save, FLAGS.seq_id_save)
# print("scan_files:{}".format(scan_files))
'''
test
print(scan_files[0])
print(scan_files_save[0])
points = load_data_points(scan_files[0])
rechange_points(points)
WriteFile(points,scan_files_save[0])
'''
for file in scan_files:
points_ = load_data_points(file)
rechange_points(points_)
points_all.append(points_)
# write bin
i = 0
for file_save in scan_files_save:
WriteFile(points_all[i],file_save)
i +=1
使用和rangenet使用数据集格式一致：
python test.py -d /path/to/dataset -s 00 -ds /path/to/save -ss 00如有错误，望指正