读取 PCD 文件时出现无效起始字节错误

Question

I'm trying to extract data from a face dataset I found online which provides png pictures and their corresponding pcd files. However, whenever I try to extract data from the pcd files I get the error:

UnicodeDecodeError: 'utf-8' codec can't decode byte 0xc0 in position 202: invalid start byte

I understand that this is because I'm trying to read a non-ASCII character, however, I haven't seen any people run into this problem when opening an outside source's.pcd files. Is there an error on the end of the dataset, or is there a workaround that will let me read this file. I eventually want to work towards a depth image for machine learning applications (I'm fairly new to machine learning in general).

If this is a problem with the dataset, I'd love to hear about other RGB-D face datasets, as I haven't been able to find any others that provide depth information.

If this is my problem, I'd like to know what I can do to fix it, because I have tried a number of different techniques and libraries to read the files and have only gotten this error.

Thanks!

import os
import math
import numpy as np
from PIL import Image


filePath = "001_01_cloud.pcd"

with open(filePath, "r") as pcd_file:
    lines = [line.strip().split(" ") for line in pcd_file.readlines()]

Answer 1

Googling for the specification for the PCD format says that the actual point cloud data could be stored in binary form, and I'm assuming that's what's going on here:

DATA - specifies the data type that the point cloud data is stored in. As of version 0.7, two data types are supported: ascii and binary.

Since you're opening the file with mode "r" , Python will assume it's text, and will handily attempt to interpret everything as UTF-8 (by default; you can pass encoding="..." ).

However since the format has a text header followed by text or binary data, you will need to open it in binary mode, "rb" . (This means reads from the file will yield bytes objects, not strings.) You can then .decode() bytes objects into strings if you need to handle them as text.

You also shouldn't use .readlines() with a file like this; the binary data that follows the textual headers can contain \n characters, and that data would be "broken" if split into lines.

Anyway, you may be reinventing the wheel here; there seems to be a Python library for PCD files .