Скачать или смотреть ShapeR: Meta’s New Way to Reconstruct 3D Scenes From Images

Meta Research released ShapeR — a 3D reconstruction system that takes a different approach to rebuilding scenes from images.

Instead of reconstructing the entire scene as one dense block, ShapeR decomposes it into individual objects and reconstructs them separately. This makes the resulting 3D representation more structured, editable, and usable for downstream tasks.

What ShapeR does:
• Reconstructs 3D scenes from multiple images
• Identifies and rebuilds individual objects, not a single fused mesh
• Produces cleaner geometry with clearer object boundaries
• Better suited for scene understanding, editing, and simulation

This object-centric approach is especially important for:
• AR / VR environments
• Robotics and manipulation
• 3D scene editing and asset reuse
• Synthetic data generation

ShapeR shows how scene reconstruction is moving away from monolithic point clouds toward semantic, object-aware 3D representations.

Links:
• Project page: https://facebookresearch.github.io/Sh...
• GitHub: https://github.com/facebookresearch/S...
• Hugging Face: https://huggingface.co/facebook/ShapeR



#ShapeR #MetaAI #3DReconstruction #ComputerVision #3DScenes #NeRF #GaussianSplatting #AR #VR #Robotics #AIResearch