问题场景 #

我在Colab中构建图像识别模型，通过挂载Google Drive导入数据集（方便他人共享时无需下载或修改路径），运行splitfolders.ratio分割数据集时触发以下错误：

OSError: [Errno 107] Transport endpoint is not connected: '/content/drive/MyDrive/dataset'

当前使用的代码：

!pip install tensorflow keras split-folders opencv-python

import splitfolders 

from google.colab import drive
drive.mount('/gdrive')
%cd /gdrive

input_folder = "/content/drive/MyDrive/dataset" 
output_folder = "/content/drive/MyDrive/image-recognition"

split_ratio = (0.8, 0.1, 0.1)

splitfolders.ratio(
    input_folder,
    output= output_folder,
    seed = 500, #para que cada ves que de dividan las imgs se haga de la misma forma
    ratio = split_ratio,
    group_prefix = None
)

错误原因 #

代码中把Google Drive挂载到了/gdrive目录，但后续引用的数据集路径却是/content/drive/MyDrive/...——两个路径不匹配，导致系统无法找到正确的挂载端点，触发传输连接错误。

解决方案 #

方案1：统一挂载路径与数据集路径 #

保持挂载到/gdrive，修改数据集路径为挂载目录下的对应路径：

!pip install tensorflow keras split-folders opencv-python

import splitfolders 

from google.colab import drive
drive.mount('/gdrive')
%cd /gdrive

# 修改路径为/gdrive下的MyDrive目录
input_folder = "/gdrive/MyDrive/dataset" 
output_folder = "/gdrive/MyDrive/image-recognition"

split_ratio = (0.8, 0.1, 0.1)

splitfolders.ratio(
    input_folder,
    output= output_folder,
    seed = 500, # 保证每次分割结果一致
    ratio = split_ratio,
    group_prefix = None
)

方案2：使用Colab默认挂载路径 #

将Drive挂载到Colab默认的/content/drive目录，保持原有数据集路径不变：

!pip install tensorflow keras split-folders opencv-python

import splitfolders 

from google.colab import drive
# 挂载到默认路径/content/drive
drive.mount('/content/drive')
%cd /content/drive/MyDrive

input_folder = "/content/drive/MyDrive/dataset" 
output_folder = "/content/drive/MyDrive/image-recognition"

split_ratio = (0.8, 0.1, 0.1)

splitfolders.ratio(
    input_folder,
    output= output_folder,
    seed = 500, # 保证每次分割结果一致
    ratio = split_ratio,
    group_prefix = None
)

额外排查步骤 #

1. 挂载完成后，先通过!ls /gdrive/MyDrive（方案1）或!ls /content/drive/MyDrive（方案2）确认dataset文件夹存在
2. 如果挂载后出现连接断开，可添加force_remount=True强制重新挂载：drive.mount('/gdrive', force_remount=True)或drive.mount('/content/drive', force_remount=True)

内容的提问来源于stack exchange，提问作者Luna