Inception-v1 (GoogLeNet) CIFAR-10

This repository contains the Inception-v1 (GoogLeNet) model trained on the CIFAR-10 dataset using PyTorch.
It achieves 91.21% test accuracy and is ready for inference or fine-tuning.

I hope you find this model useful and easy to integrate into your projects.

---

Model Description

• Architecture: Inception-v1 (GoogLeNet) with auxiliary classifiers
• Dataset: CIFAR-10 (10 classes, 60,000 32x32 color images)
• Performance: 91.21% test accuracy
• Framework: PyTorch
• Input: 3x32x32 images (RGB)
• Output: Class probabilities for 10 classes:
  airplane, automobile, bird, cat, deer, dog, frog, horse, ship, truck

---

🔗 Links

• GitHub Repository: CNN Models GitHub Repo
• Streamlit: Link

---

Usage

Feel free to download and use the model in your own projects. Here's a simple example:

import torch
from googlenet_model import GoogLeNetCIFAR  # Make sure this contains the model implementation

# Load the pretrained model
model = GoogLeNetCIFAR(num_classes=10)
model.load_state_dict(torch.load("Inception-v1.pth", map_location=torch.device('cpu')))
model.eval()

# Example inference
from torchvision import transforms
from PIL import Image

transform = transforms.Compose([
    transforms.Resize((32, 32)),
    transforms.ToTensor(),
    transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
])

image = Image.open("example_image.png")
input_tensor = transform(image).unsqueeze(0)  # Add batch dimension

output, _, _ = model(input_tensor)  # Main output and auxiliary classifiers
predicted_class = output.argmax(1).item()
print("Predicted Class:", predicted_class)

---

Training Details

• Optimizer: Adam with weight decay 5e-4
• Learning Rate Scheduler: StepLR (step=15, gamma=0.1)
• Loss Function: CrossEntropyLoss with auxiliary loss weighting
• Weight Decay: 5e-4
• Batch Size: 128
• Epochs: 50
• Data Augmentation: Random horizontal flip, normalization

---

License

This model is released under the MIT License.