使用OpenCV训练人脸识别器时仅训练单张脸的解决方法咨询

Let's troubleshoot why your code is only training on a single face. Your data collection loop looks valid on the surface, but there are a few common missteps that could be limiting it to one sample. Here's how to fix it:

1. Make sure imagePaths includes all your training images #

If imagePaths only contains a single file path, your loop will naturally only process one image. Double-check how you're generating this list—you probably need to fetch all image files in your dataset directory, not just one.

For example, use glob to grab all JPG/PNG files in your target folder:

import glob
# Update this path to match your dataset's location
imagePaths = glob.glob("path/to/your/face_dataset/*.jpg") + glob.glob("path/to/your/face_dataset/*.png")

Also confirm your filename format matches your ID extraction logic: your code expects filenames like user.1.jpg (where the ID is the second segment split by dots). If your filenames follow a different structure, adjust the split logic accordingly.

2. Initialize faces and IDs as empty lists before the loop #

If you forgot to initialize these lists, or if they were accidentally set to single values earlier in your code, the append calls won't build a collection of multiple samples. Add these lines right before your loop:

faces = []
IDs = []

3. Verify your training step uses all collected samples #

Even if you're gathering multiple face data points, if you don't pass all of them to the recognizer's training method, it will only use one. For example, if you're using OpenCV's LBPHFaceRecognizer, your training code should look like this:

import cv2
# Initialize the recognizer
recognizer = cv2.face.LBPHFaceRecognizer_create()
# Train with ALL collected faces and IDs (convert IDs to a numpy array)
recognizer.train(faces, np.array(IDs))
# Save the trained model for later use
recognizer.save("trained_face_model.yml")

Full Working Example #

Here's the complete, corrected code with debugging prints to confirm you're processing multiple images:

import os
import numpy as np
from PIL import Image
import cv2
import glob

# 1. Fetch all image paths
imagePaths = glob.glob("path/to/your/face_dataset/*.[jp][pn]g")

# 2. Initialize storage lists
faces = []
IDs = []

# 3. Process each image
for imagePath in imagePaths:
    print(f"Processing image: {imagePath}")  # Debug: confirm we're looping through multiple files
    faceImg = Image.open(imagePath).convert('L')
    faceNp = np.array(faceImg, 'uint8')
    
    # Extract ID from filename
    filename = os.path.split(imagePath)[1]
    ID = int(filename.split(".")[1])
    print(f"Extracted ID: {ID}")  # Debug: confirm correct ID extraction
    
    faces.append(faceNp)
    IDs.append(ID)

# 4. Train the recognizer
recognizer = cv2.face.LBPHFaceRecognizer_create()
recognizer.train(faces, np.array(IDs))
recognizer.save("trained_face_model.yml")

print(f"Training complete! Processed {len(faces)} samples from {len(set(IDs))} unique people.")

Run this code, and the debug prints will show you exactly which images are being processed. If you still see only one image in the output, double-check your imagePaths path—you might be pointing to the wrong directory.

内容的提问来源于stack exchange，提问作者joseph