Researchers have developed MAVE, a framework designed to reconstruct editable 3D blockouts from film shots, significantly reducing the manual effort involved. This system combines metric observation, agentic modeling, and verification to recover cameras, human motion, and geometric evidence, which are then used to construct an editable environment. MAVE improves scores on the new MAVEBench benchmark by up to 41.1 points and achieves a score of 73.2 with DeepSeek-1017 at a low API cost, while also reducing reconstruction time by 53%. The project also introduces MAVEData-3K, a dataset of 3,000 blockout-video pairs for model training, and will release the code publicly. AI
IMPACT This framework could streamline 3D content creation for film and video generation, potentially lowering production costs and increasing accessibility.
RANK_REASON This is a research paper detailing a new AI framework and benchmark. [lever_c_demoted from research: ic=1 ai=1.0]
- alphaXiv
- arXiv
- Blender
- CatalyzeX Code Finder for Papers
- Connected Papers
- DagsHub
- DeepSeek-1017
- Gotit.pub
- Hugging Face
- Litmaps
- MAVE
- MAVEBench
- MAVEData-3K
- ScienceCast
- scite Smart Citations
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →