WebUI for ML-Sharp (3DGS)

A seamless Pinokio-ready Web UI for Apple's ML-Sharp, allowing you to generate 3D Gaussian Splatting (3DGS) models from a single image with high efficiency.

🚀 Features

• Ultra-Fast 3DGS Generation: High-speed production of quality 3D Gaussian Splatting (3DGS) models from a single image.
• Single Click Install: Fully automated dependency management via Pinokio.
• Cross-Platform Rendering: Support for high-performance video rendering on both NVIDIA (CUDA) and Apple Silicon (MPS).
• Low VRAM Support: Optimized PLY generation modes for systems with limited GPU memory.
• Stable GPU Workflow: Optimized subprocess-based rendering for reliable memory release and maximum stability.
• Flexible Resolution: Dynamic resolution selector with predefined presets for optimal quality/speed balance.
• Intuitive Web UI:
  • New Job: Upload an image and get your 3D assets immediately.
  • Result History: Manage, download, and review your previous generations.
  • Graceful Cancellation: Instant "Stop" support to interrupt long-running jobs and release resources immediately.
• Advanced Camera Control: Includes 10 different camera trajectories (Orbit, Pan, Dolly, etc.).
• Smart Resource Management: Automatic GPU memory cleanup and cache management to keep your installation lean.

📦 Installation

1. Open Pinokio.
2. Click on Discover or Download.
3. Paste the URL of this repository.
4. Click Install.
5. Once finished, click Start.

🛠 Usage

1. Upload: Drag and drop an image into the "New Job" tab.
2. Configure: Choose your desired Video Resolution and select "Generate Video Immediately" if you have a supported GPU (NVIDIA or Mac M-series).
3. Run: Click "Start Generation".
4. Download: Once finished, download the .ply model or videos from the file list.
5. History: Access the "Result History" tab to manage your collection.

🖥 Requirements

• Pinokio: https://pinokio.computer

🍏 Apple Mac (Silicon)

• Chip: M1, M2, M3 or newer.
• Memory: Minimum 16GB Unified Memory is required for stable operation.
• Video Rendering: Fully supported via MPS (Metal Performance Shaders).
  • Note: Includes custom precompiled gsplat (MPS-Lite) wheels for instant setup without compilation.

🖥️ Windows / Linux PC

• GPU: NVIDIA GPU required for CUDA video rendering.
• VRAM:
  • Recommended: 10GB+ for seamless performance.
  • Minimum: 8GB supported via Low VRAM Mode (FP16) and shared memory.
• Libraries: Includes precompiled gsplat wheels for standard CUDA configurations.

⚠️ Other

• CPU Mode: Supported via generic installation (slower, video rendering disabled).

⚖️ Licenses

This software is free for Personal, Non-Commercial, and Research Use Only.

• Project License: Modified MIT (Personal/Research Usage).
• Apple ML-Sharp: Subject to Apple's Research Model License and Personal Software License.
• gsplat: Apache 2.0.

Please review the full license details in the "Licenses & Credits" tab within the application.

📜 Credits

• SHARP (ML-Sharp): Sharp Monocular View Synthesis in Less Than a Second. GitHub Repository
• 3DGS: 3D Gaussian Splatting for Real-Time Radiance Field Rendering.

---

Developed for Pinokio Ecosystem.