MOSIV: Multi-Object System Identification from Videos

Dimensions overall score 8.0

GitHub Code Pulse

No public code linked for this paper yet.

• MOSIV introduces a new framework for multi-object system identification from videos that directly optimizes for continuous, per-object material parameters.method 95%
• Prior methods are ill-suited for multi-object system identification because they focus on single-object scenes or discrete material classification.limitation 90%
• MOSIV substantially improves grounding accuracy over adapted baselines on a new synthetic benchmark.result 85%
• MOSIV substantially improves long-horizon simulation fidelity over adapted baselines on a new synthetic benchmark.result 85%
• The paper presents a new synthetic benchmark with contact-rich, multi-object interactions to facilitate evaluation of multi-object system identification.method 90%
• Object-level fine-grained supervision is critical for stable optimization in complex, multi-object system identification settings.technical 80%
• Geometry-aligned objectives are critical for stable optimization in complex, multi-object system identification settings.technical 80%
• MOSIV establishes a strong baseline for the new task of multi-object system identification from videos.result 75%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card