This model performs 4-class texture classification on cheese images using AutoML-optimized transfer learning. The model was developed by Rumi Loghmani using the cheese image dataset created by Aslan Noorghasemi.

Model Description

Architecture: Transfer Learning with resnet34
Task: 4-class texture classification (Low, Medium-Low, Medium-High, High texture)
Input: 224x224 RGB images
Output: 4-class probability distribution

Training Details

Model Developer: Rumi Loghmani (@rlogh)
Dataset: aslan-ng/cheese-image (by Aslan Noorghasemi)
AutoML Method: Optuna with 20 trials
Transfer Learning: Pre-trained resnet34 backbone
Early Stopping: Yes (patience=10)
Max Epochs: 50

Performance

Test Accuracy: 80.00%
Validation Accuracy: 50.00%
Test Loss: 1.6345

Best Hyperparameters (AutoML Optimized)

json

{
  "model_name": "resnet34",
  "dropout_rate": 0.4682019316470914,
  "learning_rate": 0.00027817005315620047,
  "weight_decay": 1.4677013775851028e-05,
  "batch_size": 2
}

Usage

python

import torch
import torch.nn as nn
from PIL import Image
import torchvision.transforms as transforms
import torchvision.models as models

# Load model (define TransferLearningModel class first)
model = TransferLearningModel(num_classes=4, dropout_rate=0.4682019316470914, model_name='resnet34')
model.load_state_dict(torch.load('pytorch_model.bin'))
model.eval()

# Preprocess image
transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])

# Load and preprocess image
image = Image.open('cheese_image.jpg').convert('RGB')
input_tensor = transform(image).unsqueeze(0)

# Make prediction
with torch.no_grad():
    output = model(input_tensor)
    probabilities = torch.softmax(output, dim=1)
    predicted_class = torch.argmax(probabilities, dim=1).item()

class_names = ["Low Texture", "Medium-Low Texture", "Medium-High Texture", "High Texture"]
print(f"Predicted class: {class_names[predicted_class]}")

Class Definitions

Class 0 (Low Texture): Texture values <= 0.425
Class 1 (Medium-Low Texture): Texture values 0.425 < x <= 0.600
Class 2 (Medium-High Texture): Texture values 0.600 < x <= 0.775
Class 3 (High Texture): Texture values > 0.775

AutoML Features

Hyperparameter Optimization: Optuna with 20 trials
Architecture Search: ResNet18 vs ResNet34
Transfer Learning: Pre-trained ImageNet weights
Early Stopping: Prevents overfitting
Fixed Budget: 20 trials, 10-minute timeout

Limitations

Trained on a very small dataset (30 images)
Texture classification may not generalize to all cheese types
Performance may vary with different lighting conditions or image quality

AI Usage

This notebook was developed with the assistance of AI as a coding co-pilot. AI tools were used to:

Suggest code snippets: AI provided suggestions for implementing various parts of the code, such as the dataset class, model architecture, training loops, and data preprocessing steps.
Debug and refactor code: AI helped identify potential errors and suggest ways to refactor code for improved readability and efficiency.
Generate documentation: AI assisted in generating explanations for code sections and creating the model card for Hugging Face.
Explore potential issues: AI provided insights into potential challenges, such as handling small batch sizes and implementing early stopping, and suggested strategies to address them.

The AI served as a valuable partner throughout the development process, accelerating coding and improving code quality.

Citation

If you use this model, please cite both the model and the original dataset:

Model Citation:

bibtex

@model{rlogh/cheese-texture-classifier-automl,
  title={Cheese Texture Classifier (AutoML)},
  author={Rumi Loghmani},
  year={2024},
  url={https://huggingface.co/rlogh/cheese-texture-classifier-automl}
}

Dataset Citation: