如何修改TuriCreate代码实现多类动物CoreML图像模型训练？

Hey there! Let's get your TuriCreate image classifier set up for all five animal classes (dog, cat, pigeon, squirrel, raccoon) without a hitch. First, let's start with the best practice for organizing your data—this will make labeling your images way simpler and less error-prone:

Step 1: Organize Your Image Files #

Create a train directory, then add 5 subdirectories inside it, each named exactly after your animal classes:

• train/dog → all dog images here
• train/cat → all cat images here
• train/pigeon → all pigeon images here
• train/squirrel → all squirrel images here
• train/raccoon → all raccoon images here

This structure lets TuriCreate easily pull the category name directly from the folder path, no messy string checks needed.

Step 2: Modified Code for 5 Classes #

Here's the updated code tailored to your five categories. I've commented each change so you know what's happening:

import turicreate as tc

# Load all images from the train directory, keeping their file paths
data = tc.image_analysis.load_images('train', with_path=True)

# Extract the category label from the folder path (e.g., "dog" from "train/dog/spot.jpg")
# Adjust the split index if your OS uses backslashes (Windows) – use path.split('\\')[-2] instead
data['label'] = data['path'].apply(lambda path: path.split('/')[-2])

# Save the labeled dataset for later use (avoids reloading images every time)
data.save('animal_images.sframe')

# Split data into 80% training, 20% testing
train_data, test_data = data.random_split(0.8)

# Train the image classifier on your labeled data
model = tc.image_classifier.create(train_data, target='label')

# Evaluate model performance on the test set (check accuracy, precision, etc.)
metrics = model.evaluate(test_data)
print("Model Evaluation Results:\n", metrics)

# Save the TuriCreate model and export to CoreML format
model.save('animal_classifier.model')
model.export_coreml('AnimalImages.mlmodel')

If You Can't Organize into Subfolders #

If your images are all in the root train folder (e.g., filenames like squirrel_001.jpg, raccoon_123.png), replace the label extraction line with this to pull the category from the filename:

import re
# Use regex to match any of your five animal classes in the file path/name
data['label'] = data['path'].apply(lambda path: re.search(r'(dog|cat|pigeon|squirrel|raccoon)', path).group())

Quick Tips for Better Results #

• Try to keep the number of images per class roughly balanced – this prevents the model from favoring over-represented classes
• TuriCreate will use GPU acceleration automatically if available, which cuts down training time significantly
• If you run into image loading errors, double-check that all files are valid JPG/PNG (TuriCreate supports most common formats)

内容的提问来源于stack exchange，提问作者Yuto