TensorFlow无法找到文件求助：读取NIST数据集图片失败

First off, I feel your pain—there’s nothing more frustrating than knowing a file exists but your code refuses to recognize it. Let’s walk through the most common fixes for this issue, using your code snippet as a starting point:

Your Current Code (for reference) #

import os
import cv2
import tensorflow as tf

upper_level_dirs = open("/Users/cam/reader/top_level_dirs")
upper_level_dirs = upper_level_dirs.read().split()
print(upper_level_dirs)

file_names = []
labe...

Top Fixes to Try #

1. Verify Paths from top_level_dirs Are Accurate

   • First, check if the paths stored in top_level_dirs are absolute paths (starting with / on Mac/Linux) or relative. If they’re relative, your code’s working directory might not match what you expect. Run print(os.getcwd()) right after importing os to confirm where your code is executing from.
   • Pick one path from upper_level_dirs, copy it into your terminal, and run ls [path] (Mac/Linux) or dir [path] (Windows) to make sure the system can find the images. Watch out for hidden spaces, typos, or mismatched capitalization—even Mac’s case-insensitive filesystem can trip up code in rare cases.

2. Use os.path.join() for Safe Path Concatenation

   • If you’re building full image paths with manual string concatenation (like dir_path + "/" + img_name), stop right now. This often leads to missing or extra path separators, which break file access. Replace it with os.path.join()—it handles system-specific separators automatically:

     for dir_path in upper_level_dirs:
         # Skip empty lines from the top_level_dirs file
         if not dir_path:
             continue
         # Iterate through files in the directory
         for img_file in os.listdir(dir_path):
             full_img_path = os.path.join(dir_path, img_file)
             # Test if cv2 can load the image
             img = cv2.imread(full_img_path)
             if img is None:
                 print(f"FAILED TO LOAD: {full_img_path}")
             else:
                 file_names.append(full_img_path)
                 # Add your label logic here
     

   • This will print exactly which paths are failing, so you can debug specific entries instead of guessing.

3. Check File Permissions

   • Even if a file exists, your user account might not have permission to read it. In the terminal, run ls -l [full_image_path] to check permissions. Look for the r flag in the user section (the first set of three characters). If it’s missing, fix it with:

     # Fix a single file
     chmod +r /path/to/your/image.png
     # Fix an entire directory recursively
     chmod -R +r /path/to/your/nist/directory
     

4. Rule Out cv2-Specific Issues

   • cv2.imread() returns None silently when it can’t find or load a file—no error message by default. To confirm it’s not a cv2 limitation, try loading a problematic image with PIL instead:

     from PIL import Image
     import traceback
     
     test_path = "path/to/a/failing/image.png"
     try:
         with Image.open(test_path) as img:
             img.verify()  # Validates the file is a proper image
         print(f"PIL successfully loaded {test_path}")
     except Exception as e:
         print(f"PIL failed to load {test_path}:")
         traceback.print_exc()
     

   • If PIL can load it, the issue is with cv2 (maybe a missing codec for the image format). If it fails too, the path or file itself is the problem.

5. TensorFlow-Specific Checks (If Using TF for Loading)

   • If you plan to use TensorFlow’s tf.data pipeline later, make sure:
     • Paths are passed as Python strings (not bytes or other types).
     • Avoid special characters (like spaces or non-ASCII characters) in paths—TF can struggle with these sometimes.
     • Use tf.io.read_file() with error handling to catch issues early:

       def load_image(path):
           try:
               img_raw = tf.io.read_file(path)
               img = tf.image.decode_png(img_raw, channels=1)  # Adjust for NIST's format
               return img
           except tf.errors.InvalidArgumentError:
               print(f"TF failed to load: {path.numpy().decode('utf-8')}")
               return None
       

Try these steps one by one—chances are the issue is a simple path typo or permission problem that’s easy to fix once you pinpoint it.

内容的提问来源于stack exchange，提问作者Cam Parra