New Segment Anything Models Make it Easier to Detect Objects and Create 3D Reconstructions
Meta has launched two powerful new AI models — SAM 3 and SAM 3D — as part of its Segment Anything Collection, significantly boosting how machines understand and manipulate visual content. SAM 3 lets you detect, segment and track objects in images and videos by using both text and visual cues. On the other hand, SAM 3D can reconstruct objects and people in full 3D from just a single image, via two versions: one for general objects and another focused on the human body. These models are now accessible through Meta’s Segment Anything Playground, designed to be user-friendly even for non-experts. Meta is also releasing the models’ weights, code, and evaluation benchmarks, enabling widespread experimentation. Potential uses range from creative content editing and AR/VR to robotics, science, and even smarter shopping experiences on Facebook Marketplace.
The Key points
- Meta introduced SAM 3 and SAM 3D, advancing its Segment Anything AI collection.
- SAM 3 supports detection, segmentation, and tracking of objects in images and videos using detailed text or visual prompts.
- It understands richer and more nuanced prompts, like “red baseball cap” or “people sitting without red hats.”
- SAM 3D includes two open-source models: one for reconstructing 3D scenes/objects (“Objects”) and another for estimating human bodies (“Body”).
- The 3D models generate geometry, texture, and layout from just a single image.
- SAM 3D Objects outperforms existing 3D reconstruction methods.
- About Facebook
- The Segment Anything Playground lets anyone try out both SAM 3 and SAM 3D — no technical expertise required.
- Meta is providing model weights, code, and new benchmarks to the community to foster research and innovation.
- These models are being applied in product features, like Facebook Marketplace’s “View in Room,” to preview furniture in your space.
- SAM 3D has potential across creative, scientific, and industrial domains — from game asset creation to robotics and sports medicine.
Disclaimer: This preview includes title, image, and description automatically sourced from the original website (about.fb.com) using publicly available metadata / OG tags. All rights, including copyright and content ownership, remain with the original publisher. If you are the content owner and wish to request removal, please contact us from your official email to no_reply@newspaperhunt.com.
