How to access images from Google Drive folders in Colab notebook

[sophiac]

I’m working on a machine learning project and need help accessing image files stored in my Google Drive from a Colab environment. My folder structure looks like this: My Drive > ML Projects > CNN_Project > petImages > training > Golden_Retriever > Golden_Retriever01.jpg. The petImages directory contains three main subdirectories (training, testing, validation). Each subdirectory has multiple folders named after different dog breeds, and each breed folder contains the actual image files. I want to load all these images from the various subdirectories and store them in a single variable called “pet_images_data” for processing. What’s the best way to mount my Google Drive and access these nested folders in Colab? Any code examples would be really helpful.

[Luke_Brilliant]

first, mount your gdrive with from google.colab import drive and then run drive.mount('/content/drive'). to get your images, try using os.listdir() or os.walk() for accessing those nested folders easily.

[DancingButterfly]

I’ve handled similar nested folder setups for computer vision projects. After mounting your drive, use the glob module - it’s perfect for pattern matching with this structure. Import glob, then run image_paths = glob.glob('/content/drive/MyDrive/ML Projects/CNN_Project/petImages/**/*.jpg', recursive=True) to grab all jpg files from every subdirectory at once. Loop through the paths to load images with opencv or PIL. The recursive parameter is crucial - it automatically traverses all those nested breed folders. Way cleaner than manually walking through each directory level, especially with multiple breeds and split folders.

[Finn_Mystery]

Try using pathlib instead - it gives you way more control over file paths. After mounting your drive, import pathlib and create a Path object for your petImages directory. Then run Path('/content/drive/MyDrive/ML Projects/CNN_Project/petImages').rglob('*.jpg') to grab all jpg files recursively. The cool thing is you can pull metadata like breed names and split types straight from the path components while loading images. I found this super helpful for building training labels - just parse the parent directory names from the path object to get breed classifications and figure out if the image goes in training, testing, or validation.