LuImage

LuImage is a convolutional neural network created from scratch with PyTorch for image classification.

Highlights

• Built from scratch with PyTorch
• No pretrained model
• Trained on CIFAR-10
• 10 image classes
• Lightweight CNN architecture

Architecture

Conv2D (3 → 16) ↓ ReLU ↓ MaxPool ↓ Conv2D (16 → 32) ↓ ReLU ↓ MaxPool ↓ Linear (2048 → 128) ↓ ReLU ↓ Linear (128 → 10)

Classes

• airplane
• automobile
• bird
• cat
• deer
• dog
• frog
• horse
• ship
• truck

Usage

import torch
from model import Luimage

model = Luimage(num_classes=10)
model.load_state_dict(
    torch.load("luimage_weights.pth", map_location="cpu")
)

model.eval()
Training
LuImage was trained from scratch using the CIFAR-10 dataset.
Optimizer: Adam
Loss: CrossEntropyLoss
Learning rate: 0.001
Batch size: 64
Epochs: 10