👟 Shoes Pair Classification

This repository contains a deep learning model for classifying pairs of shoes. The model is trained using convolutional neural networks (CNNs) to determine whether two shoes belong to the same pair.

---

📜 Table of Contents

• 🗂 Dataset
• 🛠 Installation
• 📊 Data Preprocessing
• 🧠 Model Training
• 📈 Model Evaluation
• 🚀 Using the Model
• 📸 Visualizations
• 📄 License

---

🗂 Dataset

The dataset consists of images of shoes, labeled as matching or non-matching pairs. The images are preprocessed and augmented to enhance model performance.

• Data is stored in data/ directory.
• Images are loaded and transformed using torchvision.transforms.
• Train-test split is performed before training.

---

🛠 Installation

To run this project, clone the repository and install the required dependencies:

# Clone the repository
git clone https://github.com/Moshe313/Shoes-Pair-Classification.git
cd Shoes-Pair-Classification

---

📊 Data Preprocessing

Before training, the data undergoes several preprocessing steps:

1. Image Resizing - All images are resized to a fixed dimension.
2. Normalization - Images are normalized for better training stability.
3. Augmentation - Data augmentation techniques (e.g., flipping, rotation) are applied.
4. Tensor Conversion - Images are converted to tensors for PyTorch compatibility.

Example of preprocessing:

data  = np.zeros([10, 3, 2, 224, 224, 3])
data_dic  = {}
path  = "your/data/path"
for file in glob.glob(path):
    filename = file.split("/")[-1]  # get the name of the .jpg file
    img = np.asarray(Image.open(file))  # read the image as a numpy array

    # remove the alpha channel, normalize and subtract by 0.5:
    data_dic[filename] = (img[:, :, :3] / 255) - 0.5

Images = sorted([(key, value) for key, value in data_dic.items()])
Images = [value[1] for value in Images]
for num_of_t in range(data.shape[0]):
    for num_in_t in range(3):
        for direction in range(2):
            data[num_of_t, num_in_t, direction] = Images[num_of_t * 6 + num_in_t * 2 + direction]

---

🧠 Model Training

The model is based on a Convolutional Neural Network (CNN) architecture.

• The CNN extracts features from images.
• A fully connected layer classifies the images.
• Loss function: Cross-Entropy Loss
• Optimizer: Adam optimizer
• Epochs: 10+ (Adjustable)

---

📈 Model Evaluation

The trained model is evaluated using:

• Accuracy metrics
• Confusion matrices
• Precision-Recall curves

Example visualization:

---

🚀 Using the Model

Once trained, you can test the model using the provided example.

# Create the positive and negative pairs
same = generate_same_pair(data)
same_to_predict = np.transpose(same, (0, 3, 1, 2))

correct_shown, incorrect_shown = False, False

sample = torch.Tensor(same_to_predict)
with torch.no_grad():
  sample_prediction = CNNChannel_model(sample)
sample_prediction = torch.argmax(sample_prediction, dim=1)
print(sample_prediction)

# display correctly classified:
correct_img, incorrect_img = np.ones((448,224,3)), np.ones((448,224,3))

for img in range(sample_prediction.shape[0]):
  if not correct_shown:
    if sample_prediction[img]:
      correct_img = same[img] + 0.5
  if not incorrect_shown:
    if not sample_prediction[img]:
      incorrect_img = same[img] + 0.5
  if correct_shown and incorrect_shown:
    break

percent = float(sum(sample_prediction)*100/sample_prediction.shape[0])
plt.figure(figsize=(7, 5))

plt.subplot(1, 2, 1)
plt.imshow(correct_img)
plt.title(f'Correctly classified {round(percent, 2)}%')

plt.subplot(1, 2, 2)
plt.imshow(incorrect_img)
plt.title(f'Incorrectly classified {round(100-percent, 2)}%')
plt.show()

To check accuracy :

print(f"CNNChannel accuracy: {get_accuracy(CNNChannel_model, data, batch_size=25)}")
print(f"CNN accuracy: {get_accuracy(CNN_model, data, batch_size=25)}")

---

📸 Visualizations

Here are some images from the training process:

Example Image Pairs

Image 1	Image 2	Label
		Matching
		Not Matching

---

📄 License

This project is licensed under the MIT License.

---

🚀 Enjoy classifying shoe pairs! 👟