Spaces:
Build error
Build error
| title: Insta360 3D Reconstruction | |
| emoji: π | |
| colorFrom: blue | |
| colorTo: green | |
| sdk: gradio | |
| sdk_version: 5.49.1 | |
| app_file: app.py | |
| pinned: false | |
| license: mit | |
| # π Insta360 3D Reconstruction | |
| Transform 360Β° videos into realistic 3D models using advanced depth estimation and surface reconstruction. | |
| ## β¨ Features | |
| - π¨ **Realistic Reconstruction** - Photorealistic point clouds and meshes | |
| - π― **Multi-Scale Processing** - Enhanced depth estimation | |
| - π **360Β° Support** - Works with equirectangular videos | |
| - π **Poisson Reconstruction** - Smooth, watertight meshes | |
| - π **Interactive Visualization** - Explore results in 3D | |
| - β‘ **Smart Frame Sampling** - Optimized for long videos | |
| ## π₯ How to Use | |
| 1. **Upload** your 360Β° video (MP4 format) | |
| 2. **Adjust** settings: | |
| - Target frames (auto-adjusted for long videos) | |
| - Views per frame (4-8) | |
| - Quality (low/medium/high) | |
| 3. **Process** - Click "Start Reconstruction" | |
| 4. **Download** - Get point cloud (.ply) and mesh (.obj) | |
| ## β±οΈ Processing Times | |
| | Video Length | Quality | Time | | |
| |-------------|---------|------| | |
| | 30 seconds | Medium | ~3-5 min | | |
| | 2 minutes | Medium | ~6-8 min | | |
| | 8 minutes | Medium | ~12-15 min | | |
| **Tip:** For 8-minute videos, start with Medium or Low quality for faster results! | |
| ## π§ Technical Details | |
| - **Depth Model:** Intel DPT-Large | |
| - **Surface Method:** Poisson Reconstruction | |
| - **Processing:** Multi-scale depth refinement with bilateral filtering | |
| - **Optimization:** Intelligent frame sampling for long videos | |
| ## π‘ Tips | |
| - **Long videos (>2 min):** Uses intelligent sampling every 15 seconds | |
| - **Best quality:** Use shorter clips (30-120 seconds) with High quality | |
| - **Quick preview:** Use Low quality with fewer frames | |
| - **Format:** Equirectangular 360Β° (2:1 aspect ratio) | |
| ## π Academic Project | |
| **Student:** [Your Name] | |
| **Course:** [Course Name] | |
| **Instructor:** Professor [Name] | |
| This project demonstrates advanced computer vision and 3D reconstruction techniques using deep learning for depth estimation and geometric algorithms for surface reconstruction. | |
| ## π¦ Export Formats | |
| - **.ply** - Point cloud with colors and normals | |
| - **.obj** - 3D mesh (compatible with Blender, Maya, etc.) | |
| ## β οΈ Notes | |
| - Processing longer videos requires more time and memory | |
| - GPU acceleration is used when available | |
| - For videos >5 minutes, consider using a shorter clip for testing first | |
| ## π Try It Now! | |
| Upload your 360Β° video and see the magic happen! β¨ |