akoulapure's picture
Initial Deployment: Best ViT Model
6cc8ae1 verified
Raw
History Blame Contribute Delete
5.01 kB
function AboutPage() {
return (
<div className="page">
<h2 className="section-title">About CattleAI</h2>
<p className="section-subtitle">
An AI-powered cattle breed classification system for Indian indigenous breeds.
</p>
<div className="about-grid">
<div className="card about-card">
<h3>🧠 The Models</h3>
<p>Four deep learning approaches were trained and rigorously compared:</p>
<ul>
<li>MLP Baseline β€” flatten + dense layers</li>
<li>CNN from Scratch β€” 5 conv blocks + GAP</li>
<li>ResNet50 Transfer Learning β€” ImageNet pretrained</li>
<li>ViT-B/16 Transfer Learning β€” Vision Transformer</li>
</ul>
<p style={{ marginTop: '0.75rem' }}>
The best model is selected using a weighted composite score considering F1, accuracy,
inference speed, and model size.
</p>
</div>
<div className="card about-card">
<h3>πŸ“Š The Dataset</h3>
<p>
3,056 images across 26 indigenous Indian breeds (21 cow + 5 buffalo breeds).
Stratified 70/15/15 train/val/test split.
</p>
<ul>
<li>Images resized to 224Γ—224 pixels</li>
<li>Augmentation: flip, rotation, jitter, crop</li>
<li>ImageNet normalization applied</li>
<li>Corrupt image validation at preprocessing</li>
</ul>
</div>
<div className="card about-card">
<h3>βš™οΈ Tech Stack</h3>
<ul>
<li>PyTorch 2.x + torchvision + timm</li>
<li>FastAPI backend with Pydantic schemas</li>
<li>React + Vite frontend</li>
<li>Docker containerized deployment</li>
<li>Config-driven experiments with YAML</li>
</ul>
</div>
<div className="card about-card">
<h3>🎯 Best Model Selection</h3>
<p>Weighted scoring ensures the production model balances performance and practicality:</p>
<div style={{ marginTop: '0.5rem', fontSize: '0.85rem', color: 'var(--color-text-secondary)' }}>
<div className="breed-card-detail"><span className="label">Macro F1</span><span className="value">50%</span></div>
<div className="breed-card-detail"><span className="label">Top-1 Accuracy</span><span className="value">20%</span></div>
<div className="breed-card-detail"><span className="label">Inference Latency</span><span className="value">15%</span></div>
<div className="breed-card-detail"><span className="label">Model Size</span><span className="value">10%</span></div>
<div className="breed-card-detail"><span className="label">Calibration</span><span className="value">5%</span></div>
</div>
</div>
<div className="card about-card">
<h3>🌾 For Farmers</h3>
<p>
This tool is designed for real-world agricultural use. Features include:
</p>
<ul>
<li>Camera capture for field use</li>
<li>Low-confidence warnings for uncertain predictions</li>
<li>Image quality tips for better results</li>
<li>Breed details including milk yield and primary use</li>
<li>Works offline after initial load (PWA-ready)</li>
</ul>
</div>
<div className="card about-card">
<h3>πŸ‘€ Creator</h3>
<p>
Built by Ajaya and team.
</p>
<p style={{ marginTop: '0.5rem' }}>
Source code on{' '}
<a href="https://github.com/amoghakoulapure/cattle-breed-classifier" target="_blank" rel="noreferrer">
GitHub
</a>.
</p>
<ul style={{ marginTop: '0.5rem' }}>
<li>Backend: FastAPI with PyTorch inference</li>
<li>Frontend: React + Vite</li>
<li>Training: Jupyter notebooks with shared ML package</li>
</ul>
</div>
</div>
</div>
);
}
export default AboutPage;