Introduction
The SAM3D (Segment Anything Model 3D) suite by Meta introduces a powerful workflow for 2D-to-3D reconstruction directly within Scenario. These models allow creators to extract geometric data from static photographs, transforming props and people into standardized 3D assets ready for visualization or animation.
SAM3D Objects
SAM3D Objects is a text-prompted reconstruction tool designed to generate 3D models from specific items within a static environmental image.
Prop Extraction: Describe target objects, such as a "wooden chair" or "desk lamp," and the system reconstructs those elements into 3D geometry.
Efficiency: It streamlines the creation of digital assets by extracting individual props directly from a scene, significantly reducing manual modeling time.
Precision Guidance: Users can use Point Prompts or Box Prompts to define the exact area of the image the model should focus on for reconstruction.
SAM3D Human Body
SAM3D Human Body is a specialized model that reconstructs anatomical human meshes from static images.
Standardized Geometry: It converts people in a photo into standardized 3D body geometry rather than exact photorealistic likenesses.
Flexible Output: The tool generates a combined model of a group as well as individual files for each person detected in the image.
Visualization Base Assets: These meshes serve as high-quality foundations for anatomical and architectural visualization projects.
SAM3D Align
SAM3D Align is an integration model that combines 3D body meshes from SAM3D Human Body with props generated by SAM3D Objects.
Scene Integration: It aligns characters and objects within a shared 3D space using the original image as a reference.
Cohesive Layouts: This process creates cohesive scenes where extracted assets maintain their spatial relationship, making them ready for immediate animation.
Perspective Matching: Includes a Focal Length setting to ensure the 3D scene perspective matches the original camera properties of the source image.
Technical Parameters & Settings
To achieve the highest fidelity in your 3D reconstructions, you can fine-tune several specific parameters:
Point & Box Prompts: Use these tools in SAM3D Objects to isolate the specific object you want to extract from a complex background.
Focal Length: Essential in SAM3D Align for matching the virtual 3D camera to the real-world photo, ensuring assets are not distorted.
Seed: Allows you to reproduce a specific reconstruction or test subtle variations.
Was this helpful?