README.md

---
title: DeepCAD Generator
emoji: 🔧
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 5.47.2
app_file: app.py
pinned: false
license: mit
python_version: 3.9
---

# 🔧 DeepCAD: Deep Generative Network for CAD Models

Generate Computer-Aided Design (CAD) models using deep learning! This Space implements the DeepCAD model from the **ICCV 2021** paper.

<p align="center">
  <img src="https://github.com/ChrisWu1997/DeepCAD/raw/master/teaser.png" width="600" alt="DeepCAD Teaser">
</p>

## 📖 About

**DeepCAD** is a deep generative network that can automatically generate novel, high-quality CAD models. This Space provides an interactive demo where you can:

- 🎲 Generate random CAD models
- 🎯 Control generation with seed values for reproducibility
- 🌡️ Adjust temperature to control creativity
- 📊 View 2D visualizations of generated models
- 💾 Download H5 files for further processing

## 🚀 How to Use

1. **Set Random Seed** (0-10000): Controls which random model is generated
2. **Adjust Temperature** (0.1-2.0): 
   - Lower values (0.5-0.8) = more conservative, structured designs
   - Higher values (1.2-2.0) = more creative, varied designs
3. **Click "Generate CAD Model"** to create a new design
4. **Download the H5 file** to use with full DeepCAD tools

## 📊 What You'll See

The visualization shows:
- **2D Projection**: Simplified view of CAD operations
- **Command Distribution**: Histogram of operation types
- **Parameter Statistics**: Key metrics about the generated model

**Note**: This is a 2D simplified visualization. For full 3D STEP file export, download the H5 file and use the [original DeepCAD repository tools](https://github.com/ChrisWu1997/DeepCAD).

## 🔬 Technical Details

### Model Architecture
- **Latent Dimension**: 256
- **Commands**: 60 CAD operations per sequence
- **Arguments**: 256 parameters per command
- **Device**: Automatic CPU/GPU detection

### Generation Process
1. Sample a random latent vector (controlled by seed)
2. Scale by temperature for variation
3. Decode through neural network
4. Output vectorized CAD sequence

## 💾 Using the Generated H5 Files

The H5 files are compatible with the original DeepCAD tools:
```bash