✨MASS: Motion-Aware Spatial-Temporal Grounding for Physics Reasoning and Comprehension in Vision-Language Models

📝 Summary:
VLMs struggle with physics-driven video reasoning. This paper introduces MASS, a method that injects spatial-temporal signals and motion tracking into VLMs, along with the MASS-Bench dataset. MASS significantly improves VLM performance on physics tasks, outperforming baselines and achieving state...

🔹 Publication Date: Published on Nov 23

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.18373
• PDF: https://arxiv.org/pdf/2511.18373

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#VLMs #PhysicsAI #ComputerVision #AIResearch #MachineLearning

✨Seeing the Wind from a Falling Leaf

📝 Summary:
This paper presents an end-to-end differentiable inverse graphics framework that recovers invisible force representations from video observations. This innovation enables estimating physical forces, like wind from a falling leaf, leading to physics-based video generation and editing.

🔹 Publication Date: Published on Nov 30

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.00762
• PDF: https://arxiv.org/pdf/2512.00762

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#InverseGraphics #PhysicsAI #ComputerVision #VideoGeneration #DeepLearning

✨ProPhy: Progressive Physical Alignment for Dynamic World Simulation

📝 Summary:
ProPhy is a two-stage framework that enhances video generation by explicitly incorporating physics-aware conditioning and anisotropic generation. It uses a Mixture-of-Physics-Experts mechanism to extract fine-grained physical priors, improving physical consistency and realism in dynamic world sim...

🔹 Publication Date: Published on Dec 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.05564
• PDF: https://arxiv.org/pdf/2512.05564

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#VideoGeneration #PhysicsAI #DynamicSimulation #DeepLearning #ComputerVision

✨PhysRVG: Physics-Aware Unified Reinforcement Learning for Video Generative Models

📝 Summary:
Existing video generation models lack physical realism, especially for rigid body collisions, treating physics rules as soft conditions. This paper introduces PhysRVG, a physics-aware reinforcement learning paradigm that strictly enforces physical collision rules directly in high-dimensional spac...

🔹 Publication Date: Published on Jan 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.11087
• PDF: https://arxiv.org/pdf/2601.11087

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#VideoGeneration #PhysicsAI #ReinforcementLearning #GenerativeAI #ComputerVision

✨PhyRPR: Training-Free Physics-Constrained Video Generation

📝 Summary:
PhyRPR introduces a three-stage pipeline Reason-Plan-Refine for video generation. It decouples physical understanding from visual synthesis, addressing issues with physical plausibility. This improves motion controllability and allows for explicit physical control during generation.

🔹 Publication Date: Published on Jan 14

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.09255
• PDF: https://arxiv.org/pdf/2601.09255

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#VideoGeneration #PhysicsAI #AIResearch #ComputerVision #DeepLearning

✨From Statics to Dynamics: Physics-Aware Image Editing with Latent Transition Priors

📝 Summary:
PhysicEdit addresses physically implausible image editing by modeling edits as predictive physical state transitions. It uses a dual-thinking diffusion framework guided by a vision-language model, greatly enhancing physical realism.

🔹 Publication Date: Published on Feb 25

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.21778
• PDF: https://arxiv.org/pdf/2602.21778
• Project Page: https://liangbingzhao.github.io/statics2dynamics/
• Github: https://github.com/liangbingzhao/PhysicEdit

✨ Datasets citing this paper:
• https://huggingface.co/datasets/metazlb/PhysicTran38K

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#ImageEditing #DiffusionModels #ComputerVision #PhysicsAI #AIResearch