We’re thrilled to introduce SAM 3D, a groundbreaking advancement in 3D reconstruction technology that transforms ordinary 2D images into detailed 3D models. This revolutionary release includes two powerful models: SAM 3D Objects for comprehensive object and scene reconstruction, and SAM 3D Body for precise human pose and shape estimation.
Experience SAM 3D Through the Segment Anything Playground
We’ve created the Segment Anything Playground, a user-friendly platform that allows anyone to experiment with these cutting-edge models. You can upload your own images, select objects or humans, and generate detailed 3D reconstructions in real-time. The platform also features SAM 3, our latest foundation model for enhanced image and video understanding.
Already, these technologies are making waves in practical applications. SAM 3D and SAM 3 power Facebook Marketplace’s new “View in Room” feature, helping users visualize how home decor items will look in their spaces before making purchases.
SAM 3D Objects: Revolutionary Object Reconstruction
SAM 3D Objects represents a paradigm shift in 3D reconstruction from single images. It tackles the complex challenge of reconstructing detailed 3D shapes, textures, and object layouts from everyday photographs, even when dealing with small objects, indirect views, or occlusions.
The breakthrough comes from our innovative data annotation engine that overcomes the traditional limitations of 3D data collection. By leveraging human expertise in verifying and ranking 3D meshes rather than creating them from scratch, we’ve annotated nearly 1 million distinct images and generated approximately 3.14 million model-in-the-loop meshes at unprecedented scale.
We’ve also introduced the SAM 3D Artist Objects dataset (SA-3DAO), a first-of-its-kind evaluation benchmark that pushes the field toward more realistic physical world 3D perception. The results speak for themselves: SAM 3D Objects achieves at least a 5:1 win rate over other leading models in human preference tests and can generate full textured reconstructions within seconds.
SAM 3D Body: Advanced Human Reconstruction
SAM 3D Body addresses the complex challenge of estimating 3D human pose and shape from single images, even in difficult scenarios involving unusual postures, occlusions, or multiple people. The model is designed to be promptable, accepting interactive inputs like segmentation masks and 2D key points for enhanced user control.
![Human Reconstruction](https://video-hkg1-2.xx.fbcdn.net/o1/v/t2/f2/m366/AQNZrjnKH__FexXFWYn4NY9UMshN8cn4VHFd45-tG77lX0TKrq8X8N_ubCpbeGWLoJSQh3Vqx1aOP3FfdEkg3ymw-IkuwCtn58rT6LPConxieQ.mp4?nc_cat=107&_nc_sid=5e9851&_nc_ht=video-hkg1-2.xx.fbcdn.net&_nc_ohc=nUnEJPedw4sQ7kNvwH67Rdy&efg=eyJ2ZW5jb2RlX3RhZyI6Inhwdl9wcm9ncmVzc2l2ZS5GQUNFQk9PSy4uQzMuMTI4MC5kYXNoX2gyNjQtYmFzaWMtZ2VuMl83MjBwIiwieHB2X2Fzc2V0X2lkIjoxNTc3NDczNzEwMjgwMzU1LCJhc3NldF9hZ2VfZGF5cyI6OSwidmlfdXNlY2FzZV9pZCI6MTAxMjgsImR1cmF0aW9uX3MiOjIxLCJ1cmxnZW5fc291cmNlIjoid3d3In0%3D&ccb=17-1&vs=4b3ba7702f7bb2c6&_nc_vs=HBkcFQIYRWZiX2VwaGVtZXJhbC9FQTRGMDUyMkFFRDVBMkE1QkNBOUYyRDBGQjdCNTlCMF9tdF8xX3ZpZGVvX2Rhc2hpbml0Lm1wNBUAAsgBEgAoABgAGwKIB3VzZV9vaWwBMRJwcm9ncmVzc2l2ZV9yZWNpcGUBMRUAACbGuqaCh63NBRUCKAJDMywXQDV1P3ztkWgYGWRhc2hfaDI2NC1iYXNpYy1nZW4yXzcyMHARAHUCZaCeAQA&_nc_gid=Su3dJvmZ9-YGRRXVBB49Rg&