Скачать или смотреть V-DPM: Real 4D Video Reconstruction with Dynamic Point Maps

V-DPM (Video Dynamic Point Maps) is a research project from Oxford VGG that tackles one of the hardest problems in vision:
reconstructing fully dynamic 4D scenes from regular video.

Instead of static meshes or NeRF-style volumes, V-DPM represents scenes as dynamic point maps (DPMs) — point clouds that move, deform, and evolve over time.

What this enables:
• True 4D reconstruction (3D + time) from monocular video
• Accurate modeling of dynamic objects and scenes
• Better handling of motion, deformation, and topology changes
• A clean separation between geometry and dynamics

This approach avoids many artifacts common in voxel or implicit-field methods and produces temporally consistent reconstructions even in complex motion.

Links:
• Project page: https://www.robots.ox.ac.uk/~vgg/rese...
• GitHub: https://github.com/eldar/vdpm
• Live demo: https://huggingface.co/spaces/edgarsu...

If you’re interested in 3D reconstruction, dynamic scenes, neural rendering, or the future of spatial AI — this one is worth a look.



#4DReconstruction #ComputerVision #DynamicScenes #PointCloud #NeuralRendering #VGG #Oxford #AIResearch #3DVision #VideoTo3D #Geometry #SpatialAI #MachineLearning #DeepLearning