TL;DR
Revelium Studio hosts a web demo that transforms a single image into a navigable 3D Gaussian Splat with depth. The tool accepts JPG, PNG and HEIC files and runs Apple's SHARP research model under non‑commercial and open‑source terms; outputs are experimental and provided 'as is'.
What happened
Revelium Studio is hosting an online demo that takes one uploaded image and converts it into a navigable 3D Gaussian Splat with depth. Users can drop a file or browse to upload images in JPG, PNG or HEIC format. Behind the demo, the site cites Apple’s SHARP machine‑learning research model (apple/ml-sharp) and notes that the demo operates under Apple's research model and open‑source licensing. The page stresses that the service is intended strictly for non‑commercial research and demonstration use, that results are experimental and offered without guarantees, and that users are responsible for how they use outputs. Uploading an image requires confirming you hold the necessary rights and that the content does not infringe laws or third‑party rights. The page also points out that “Apple” and “SHARP” may be trademarks of Apple Inc.
Why it matters
- A single‑image pipeline to produce a navigable 3D representation could lower the barrier to exploring 3D content from ordinary photos.
- Making a research model available as a browser demo lets researchers and hobbyists test behavior without local setup.
- The non‑commercial and experimental restrictions mean this is targeted at research and demonstration rather than production use.
- The responsibility clause highlights legal and copyright risks for people uploading third‑party or protected imagery.
Key facts
- Demo hosted at lab.revelium.studio/ml-sharp.
- Interface invites users to drop an image or click to browse.
- Supported input file types listed: JPG, PNG, HEIC.
- Demo cites Apple’s SHARP machine‑learning research model (apple/ml-sharp).
- Operates under Apple’s research model and open‑source licenses.
- Available only for non‑commercial research and demonstration purposes.
- Outputs are described as experimental and provided “as is.”
- Users must confirm they hold rights to uploaded content and that it does not violate laws or third‑party rights.
What to watch next
- Whether the demo’s licensing or terms change to permit commercial use — not confirmed in the source.
- Any published benchmarks or technical notes on fidelity, performance, and limitations of the SHARP model via this demo — not confirmed in the source.
- If Revelium Studio or Apple publish follow‑up research or a more formal release of tools or APIs — not confirmed in the source.
Quick glossary
- Gaussian splat: A rendering primitive that represents a point in 3D space as a blurred or soft‑edged kernel, often used to approximate continuous surfaces from point samples.
- Depth map: An image where each pixel encodes distance from the camera, used to reconstruct or infer three‑dimensional structure from a two‑dimensional view.
- SHARP (apple/ml-sharp): A named machine‑learning research model referenced by the demo; identified on the page as Apple’s SHARP model (apple/ml-sharp).
- Research/open‑source license: A licensing approach that permits use for research and demonstration under specified conditions but may restrict commercial exploitation.
Reader FAQ
What image formats can I upload?
The demo lists JPG, PNG and HEIC as supported formats.
Can I use the demo output commercially?
The page states the demo is provided only for non‑commercial research and demonstration purposes.
Is the output production‑ready and guaranteed?
No — the site describes outputs as experimental and provides them “as is.”
Do I retain responsibility for uploaded images?
Yes. The site requires you to confirm you hold necessary rights and that content does not violate laws or third‑party rights.
ML-Sharp Drop your image here or click to browse • JPG, PNG, HEIC supported Disclaimer. This demo uses Apple's SHARP machine‑learning research model (apple/ml-sharp) under Apple's research model and open‑source…
Sources
- Turn a single image into a navigable 3D Gaussian Splat with depth
- Turn a single image into a navigable 3D Gaussian Splat …
- Personalized Real-to-Sim-to-Real Navigation with Gaussian …
- From Images to Semantic 3D Gaussian Splatting with Python
Related posts
- Most developers don’t trust AI-generated code but often skip verification
- AI Zealotry: A Senior Engineer’s Practical Guide to Developing with AI
- How Ozlo Is Turning Sleepbuds into a Data Platform for Sleep