Stable Diffusion Reddit1d ago|研究・論文プロダクト・サービス

ComfyUI-Sharp — Monocular 3DGS Under 1 Second via Apple's SHARP Model

A new release of ComfyUI-Sharp, which uses Apple's SHARP model to generate 3D Gaussians from a single image in under 10 seconds on CPU/GPU.

💡

Why it matters

This release demonstrates the potential of Apple's SHARP model to enable fast, monocular 3D geometry sensing, which could have applications in various industries.

Key Points

  • 1Monocular 3D Gaussian generation from a single image
  • 2Very fast inference time, less than 10 seconds on CPU/GPU
  • 3Auto focal length extraction from EXIF metadata

Details

ComfyUI-Sharp is a new release that integrates Apple's SHARP (Simultaneous Hierarchical Aggregation and Refinement for 3D Geometry Sensing) model, which can generate 3D Gaussians from a single input image. The model is able to perform this task in under 10 seconds on CPU, MPS, or GPU. The release also includes functionality to automatically extract the focal length from the image's EXIF metadata, eliminating the need for manual input. The project provides two example workflows, one with manual focal length and one with EXIF auto-extraction. The developer is seeking feedback on the model's performance with different image types and compositions, as well as its integration with downstream 3D Gaussian viewers and tools.

Like
Save
Read original
Cached
Comments
?

No comments yet

Be the first to comment

AI Curator - Daily AI News Curation

AI Curator

Your AI news assistant

Ask me anything about AI

I can help you understand AI news, trends, and technologies