SAM 3D: Transforming Single Images into High-Fidelity 3D Reality
Meta introduces the next evolution in the Segment Anything Model family. SAM 3D brings human-level 3D perception to computer vision, enabling robust reconstruction of objects and human bodies from single 'in-the-wild' images.
Powered by the new SAM 3D Objects and SAM 3D Body models, we are bridging the gap between 2D imagery and 3D spatial understanding with unprecedented accuracy and speed.
🔥 State-of-the-art performance on physical world images
Two Breakthrough Models for Comprehensive 3D Understanding
SAM 3D introduces two specialized architectures designed to handle the complexities of the physical world. Moving beyond synthetic datasets, these models leverage a massive new data engine to understand occlusion, diverse lighting, and complex geometries.
SAM 3D Objects: Scene-Aware Reconstruction
Reconstruct detailed 3D shapes, textures, and layouts from a single image. Unlike previous methods limited to isolated objects, SAM 3D Objects thrives in cluttered, real-world environments. It allows users to select any object in a photo and instantly generate a pose-aware 3D mesh, enabling applications like 'View in Room' for e-commerce and interactive 3D scene editing.
SAM 3D Body: Precise Human Digitization
Generate accurate 3D human pose and shape estimations even in challenging conditions involving unusual postures or partial occlusions. Built on the new Meta Momentum Human Rig (MHR), it separates skeletal structure from soft tissue shape, offering interpretable and animatable 3D human models for VR, gaming, and sports analysis.
The Data Engine Advantage
Our revolutionary 'Human-in-the-Loop' data engine annotated nearly 1 million physical world images, generating over 3 million verified meshes. This approach bridges the 'sim-to-real' gap, allowing SAM 3D to generalize across diverse real-world scenarios where purely synthetic-trained models fail.
Why SAM 3D Redefines the Standard
We've moved beyond the limitations of traditional photogrammetry and synthetic-only training. SAM 3D brings semantic understanding to 3D reconstruction.
From 2D Image to 3D Asset in 4 Steps
Experience the power of SAM 3D through our intuitive Playground or integrate it directly into your Python workflow.
Upload Your Image
Start with any standard RGB image. It could be a photo from your phone, a frame from a video, or a generated image.
Prompt the Model
Click on the object or person you want to reconstruct. SAM 3D instantly generates a segmentation mask to confirm your selection.
Generate 3D Mesh
The model infers the 3D geometry, texture, and pose, producing a high-quality mesh in seconds using our efficient inference pipeline.
Export & Integrate
Download your asset in standard formats like .OBJ, .GLB, or the new MHR format for human bodies, ready for use in Blender, Unity, or Unreal Engine.
Technical Specifications & Capabilities
A deep dive into the technical prowess that makes SAM 3D the new benchmark for 3D reconstruction.
Open Vocabulary Support
Reconstruct any object category without limitation, thanks to large-scale pre-training on diverse concepts.
Meta Momentum Human Rig (MHR)
A new open-source mesh format separating skeletal structure from soft tissue for superior animation.
Occlusion Handling
Advanced inference logic hallucinates plausible back-facing geometry for occluded objects.
SA-3DAO Dataset
Trained on the new SAM 3D Artist Objects dataset, featuring 1M+ visually grounded real-world images.
Multi-View Consistency
Ensures texture and geometric consistency across different viewing angles of the reconstructed object.
Apache 2.0 License
Fully open-source code and model weights, enabling commercial usage and community innovation.
Frequently Asked Questions
Common questions about SAM 3D capabilities, licensing, and usage.
Start Building in 3D Today
Join the community of developers pushing the boundaries of 3D computer vision.