docs: Revise README to focus on project overview and demo, removing outdated sections

Author: harishy0406

README.md

@@ -1,303 +1,49 @@
-# ✂️ PruneVision AI — Self-Pruning Neural Networks
+# ✂️ PruneVision AI — Short Project Report
 
-[![Python 3.9+](https://img.shields.io/badge/Python-3.9+-blue.svg)](https://www.python.org/downloads/)
-[![Docker](https://img.shields.io/badge/Docker-Ready-blue)](https://www.docker.com/)
-[![Streamlit](https://img.shields.io/badge/Streamlit-1.30%2B-FF4B4B)](https://streamlit.io/)
-[![License MIT](https://img.shields.io/badge/License-MIT-green.svg)](LICENSE)
-[![Tests](https://img.shields.io/badge/tests-passing-brightgreen)](https://github.com/prunevision/ai/actions)
+PruneVision AI is a PyTorch-based self-pruning vision framework built around learnable sigmoid gates. The project trains compact classifiers on CIFAR-10, applies L1 regularization to encourage sparse gates, and exports pruned models for efficient inference. The implementation is organized under [Full Implementation](Full%20Implementation), while the repository root stays focused on this report.
 
-**PruneVision AI** is a production-grade platform for self-pruning neural networks, enabling efficient edge deployment with minimal accuracy loss. Built with learnable gating mechanisms, it automates model compression during training.
+## Overview
 
-### The Problem
-Modern neural networks achieve exceptional accuracy but at significant computational cost:
-- ❌ 100M-1B+ parameters require 4-12GB memory
-- ❌ 100-500ms inference latency on CPU
-- ❌ 10-50W power consumption
-- ❌ Weeks of expert-driven post-training optimization
+The codebase is centered on three layers:
 
-### Our Solution
-PruneVision introduces **self-pruning architecture** with learnable gating mechanisms:
-- ✅ **50-80% parameter reduction** during training
-- ✅ **3-5× faster inference** on edge devices (ARM, NVIDIA Jetson)
-- ✅ **≤2% accuracy loss** vs. baseline
-- ✅ **Production-ready** without post-training optimization
+- **Modeling**: prunable MobileNetV3, ResNet-18, and EfficientNet-B0 backbones
+- **Training**: a staged sparsity schedule that balances accuracy and compression
+- **Deployment**: evaluation, pruning, ONNX export, and dashboard visualization
 
-### Real-World Impact
-Perfect for lightweight image classification and edge AI research:
-- 🧠 **CIFAR-10 Benchmarking** - Compare pruning across compact vision models
-- ⚡ **Edge Deployment** - Efficient inference on resource-constrained devices
-- 📉 **Model Compression** - Reduce parameters with minimal accuracy loss
-- 🔬 **Research Prototyping** - Study sparsity, gating, and transfer learning
+The best saved CIFAR-10 checkpoint in the repository is the MobileNetV3-Small run, which reports **92.37% test accuracy** and **45.3% sparsity**.
 
-## ✨ Key Features
+## Answers
 
-### 🧠 Advanced Architecture
-```python
-# Learnable gate parameters for automatic pruning
-masked_weight = weight × sigmoid(gate)
-loss = CrossEntropy + λ × L1(sigmoid(gates))
-```
+### Why L1 on sigmoid gates encourages sparsity
 
-### 🎯 3-Stage Training Pipeline
-| Stage | Epochs | Lambda | Purpose |
-|-------|--------|--------|---------|
-| **Warm-up** | 1-10 | 0.0001 | Network learns representations |
-| **Progressive** | 11-25 | 0.001→0.01 | Aggressive gate pruning |
-| **Fine-tuning** | 26-30 | 0.0 | Accuracy recovery |
+Each gate is passed through a sigmoid, so its value stays in the range $[0, 1]$. Adding an L1 penalty on those gate values makes the optimizer pay a cost for keeping gates open. The easiest way to reduce that cost is to push unimportant gates toward 0, which shuts off channels and produces a sparse model.
 
-### 📊 Interactive Dashboard
-- Real-time training metrics and visualization
-- Advanced model analysis and layer inspection
-- Inference benchmarking and comparison
-- One-click model export (ONNX, PyTorch)
+### Result summary by $\lambda$
 
-### 🐳 Production-Ready
-- Docker support with multi-stage builds
-- Kubernetes manifests included
-- Cloud deployment templates (AWS, GCP, Azure)
-- Comprehensive security hardening
+Only one completed run is stored in the current workspace, so the table below reflects the saved best MobileNetV3-Small checkpoint.
 
-### 🔒 Security & Quality
-- Type hints throughout (PEP 484)
-- 80%+ test coverage with pytest
-- Static analysis (mypy, pylint, flake8)
-- Security scanning (Bandit, Dependabot)
-- Pre-commit hooks for code quality
+| Lambda | Test Accuracy | Sparsity Level (%) |
+|-------:|--------------:|-------------------:|
+| 0.0001 | 92.37 | 45.3 |
 
+### Final gate distribution plot
 
-## 📊 Quick Comparison
 
-| Aspect | Baseline | PruneVision | Improvement |
-|--------|----------|-------------|-------------|
-| Parameters | 2.5M | 1.2M | **52% smaller** |
-| Model Size | 9.8MB | 4.7MB | **52% reduction** |
-| Inference Time | 124ms | 78ms | **37% faster** |
-| Accuracy | 89.2% | 88.9% | **-0.3%** |
 
-## 📊 Model Sparsity Comparison
+## Snapshots and Demo
 
-PruneVision AI achieves significant parameter reduction across different architectures on CIFAR-10:
+### Dashboard snapshots
 
-| Model | Baseline Params | Pruned Params | Sparsity | Accuracy Loss | Inference Speedup |
-|-------|----------------|---------------|----------|---------------|-------------------|
-| **MobileNetV3-Small** | 2.5M | 900K | **63.9%** | <1.5% | 2.8× |
-| **ResNet-18** | 11.7M | 4.2M | 64.1% | <1.2% | 2.9× |
-| **EfficientNet-B0** | 5.3M | 1.9M | 64.3% | <1.0% | 3.1× |
+![Overview snapshot](https://github.com/user-attachments/assets/81232db9-f76a-4e30-8c89-2215fd389ad2)
 
-**Key Insights:**
-- All models achieve **60%+ sparsity** with minimal accuracy loss
-- **EfficientNet-B0** shows highest compression ratio (64.3%)
-- **MobileNetV3-Small** provides best balance of sparsity and speed
-- Self-pruning gates enable automatic parameter removal during training
+![Dataset explorer snapshot](https://github.com/user-attachments/assets/186c029e-676f-4d43-9a99-36e89d09b737)
 
-*Benchmarks on CIFAR-10 dataset (50k training, 10k test images)*
+![Training monitor snapshot](https://github.com/user-attachments/assets/fda92861-cf99-46bf-a9bf-722bf758aa63)
 
-## 🚀 Quick Start
+![Model analysis snapshot](https://github.com/user-attachments/assets/c444d36f-0ae1-4ef1-97bc-0fb7859cb1ee)
 
-### 1. Installation
+![Deployment snapshot](https://github.com/user-attachments/assets/c6b0e285-7961-4d1e-a303-a541a6b983f8)
 
-```bash
-# Clone repository
-git clone https://github.com/yourusername/PruneVision-AI.git
-cd PruneVision-AI
-
-# Create virtual environment
-python -m venv venv
-source venv/bin/activate  # On Windows: venv\Scripts\activate
-
-# Install dependencies
-pip install -r requirements.txt
-```
-
-### 2. Train a Model
-
-```bash
-# MobileNetV3-Small (fastest, recommended for CPU)
-python train_model.py --model mobilenetv3_small --epochs 30
-
-# ResNet-18
-python train_model.py --model resnet18 --epochs 30
-
-# EfficientNet-B0
-python train_model.py --model efficientnet_b0 --epochs 30
-
-# With ONNX export
-python train_model.py --model mobilenetv3_small --export-onnx
-```
-
-### Launch Dashboard
-```bash
-streamlit run app.py
-```
-
-## 📊 Dataset
-
-- **CIFAR-10 Dataset**: 10 object classification classes
-  - Airplane, Automobile, Bird, Cat, Deer, Dog, Frog, Horse, Ship, Truck
-- **60,000 total images** (32×32 RGB, auto-downloaded from torchvision)
-  - Training: 50,000 images (5,000 per class)
-  - Test: 10,000 images (1,000 per class)
-- Balanced dataset with equal samples per class
-- Class-weighted sampling for optimal training convergence
-
-## 🏗️ Model Zoo
-
-| Model | Params | Size | Target Use |
-|-------|--------|------|-----------|
-| MobileNetV3-Small | ~2.5M | ~10 MB | Ultra-fast edge inference |
-| ResNet-18 | ~11.7M | ~45 MB | Balanced accuracy/speed |
-| EfficientNet-B0 | ~5.3M | ~20 MB | Best accuracy/efficiency |
-
-## 🔧 Configuration
-
-All hyperparameters are centralized in `config.py`:
-- Image size, batch size, learning rate
-- 3-stage sparsity schedule (λ values, epoch boundaries)
-- Gate initialization bias, pruning threshold
-- Data augmentation parameters
-
-## 📈 Dashboard Features
-
-1. **Overview** — Architecture explanation, pipeline visualization
-2. **Dataset Explorer** — Class distribution, sample image browser
-3. **Training Monitor** — Loss/accuracy/sparsity curves, λ schedule
-4. **Model Analysis** — Layer-wise sparsity, gate distributions
-5. **Deployment** — Model comparison, export commands
-6. **Live Demo** — Upload image → real-time classification
-
-
-## 🏗️ Architecture
-
-### System Design
-
-<img width="6531" height="3175" alt="image" src="https://github.com/user-attachments/assets/e44761a8-3663-4afb-863a-e9f44f786f31" />
-
-
-## 📈 Performance Metrics
-
-### Benchmark Results
-
-| Model | Baseline Params | Pruned Params | Reduction | Top-1 Acc | Latency (CPU) | Model Size |
-|-------|-----------------|---------------|-----------|-----------|---------------|------------|
-| **MobileNetV3-Small** | 2.5M | 1.2M | 52% ↓ | 84.2% | 45ms | 4.8MB |
-| **ResNet-18** | 11.7M | 5.2M | 55% ↓ | 89.5% | 120ms | 20.1MB |
-| **EfficientNet-B0** | 5.3M | 2.1M | 60% ↓ | 91.2% | 80ms | 8.4MB |
-
-### Edge Device Performance
-
-**Latency (ms) - Lower is Better**
-
-```
-                    Baseline    Pruned      Speedup
-┌─────────────────────────────────────────────────┐
-│ Raspberry Pi 4                                  │
-│ ████████████████ 250ms  ████ 45ms       5.5×    │
-│                                                 │
-│ NVIDIA Jetson Nano                              │
-│ ████████████ 180ms     ██ 35ms         5.1×     │
-│                                                 │
-│ Intel Core i7 (CPU)                             │
-│ ████████ 120ms        ██ 22ms          5.4×     │
-└─────────────────────────────────────────────────┘
-```
-
-### Memory Usage Comparison
-
-| Device | Baseline | Pruned | Reduction |
-|--------|----------|--------|-----------|
-| Raspberry Pi 4 (1GB RAM) | ✗ OOM | ✓ 312MB | 70% ↓ |
-| Jetson Nano (4GB RAM) | 2.1GB | 0.6GB | 71% ↓ |
-| Desktop (16GB RAM) | 4.2GB | 1.2GB | 71% ↓ |
-
-### Accuracy Preservation
-
-Models maintain competitive accuracy with minimal loss:
-
-```
-MobileNetV3-Small:  84.2% → 82.8%  (-1.4%)
-ResNet-18:          89.5% → 88.6%  (-0.9%)
-EfficientNet-B0:    91.2% → 90.4%  (-0.8%)
-```
----
-
-
-## 📊 Dashboard 
-
-### Interactive Web Interface
-
-Access comprehensive monitoring and analysis:
-
-```bash
-streamlit run app.py
-# Visit: http://localhost:8501
-```
-
-### Dashboard Tabs
-#### **Live Demo**
-- Image upload and classification
-- Real-time inference
-- Prediction confidence scores
-![image](https://github.com/harishy0406/PruneVision-AI/blob/main/Work%20Demo(Prunvison%20AI).gif)
-
-#### 1️⃣ **Overview**
-- Architecture explanation
-- Training pipeline visualization
-- Performance comparison table
-<img width="1915" height="787" alt="image" src="https://github.com/user-attachments/assets/81232db9-f76a-4e30-8c89-2215fd389ad2" />
-
-
-
-#### 2️⃣ **Dataset Explorer**
-- Class distribution histogram
-- Sample image browser
-- Dataset statistics
-<img width="1914" height="788" alt="image" src="https://github.com/user-attachments/assets/186c029e-676f-4d43-9a99-36e89d09b737" />
-
-
-
-#### 3️⃣ **Training Monitor**
-- Real-time loss and accuracy curves
-- Sparsity progression
-- Lambda schedule visualization
-<img width="1910" height="794" alt="image" src="https://github.com/user-attachments/assets/fda92861-cf99-46bf-a9bf-722bf758aa63" />
-
-
-#### 4️⃣ **Model Analysis**
-- Layer-wise sparsity heatmap
-- Gate value distribution
-- Parameter reduction statistics
-<img width="1895" height="792" alt="image" src="https://github.com/user-attachments/assets/c444d36f-0ae1-4ef1-97bc-0fb7859cb1ee" />
-
-
-#### 5️⃣ **Deployment**
-- Model comparison table
-- Export options (ONNX, PyTorch)
-- Cloud deployment guides
-<img width="1867" height="787" alt="image" src="https://github.com/user-attachments/assets/c6b0e285-7961-4d1e-a303-a541a6b983f8" />
-
-
-
-
----
-## 📜 License
-
-This project is licensed under the **Apache License 2.0**.
-
-### Third-party Licenses
-
-- PyTorch: BSD-3-Clause
-- Streamlit: Apache 2.0
-- NumPy: BSD-3-Clause
-
----
-
-<div align="center">
-
-**Made with ❤️ by M Harish Gautham**
-
-⭐ If you find this project helpful, please star it! ⭐
-
-[Website](https://github.com/harishy0406/PruneVision-AI) • [Docs](https://github.com/harishy0406/PruneVision-AI) • [GitHub](https://github.com/harishy0406/PruneVision-AI)
-
-</div>
+### Project demo
 
+[Watch the demo GIF](Full%20Implementation/Work%20Demo(Prunvison%20AI).gif)