Spring-Gaus is a novel method for reconstructing and simulating elastic objects from multi-view videos. It integrates a 3D Spring-Mass model with 3D Gaussians to capture both the visual appearance and physical dynamics of elastic objects. The method enables accurate reconstruction and simulation of elastic objects, allowing for future prediction and simulation under various initial states and environmental conditions. The core challenge is developing an expressive and efficient physical dynamics model that can capture the complex behaviors of elastic objects, including collisions, deformations, and bouncing. Spring-Gaus uses a learnable system of mass points and springs to model the physical properties of elastic objects. It decouples the appearance and geometry reconstruction from the physical dynamics reconstruction, enabling more effective optimization. The method is evaluated on both synthetic and real-world datasets, demonstrating its ability to accurately reconstruct and simulate elastic objects. Spring-Gaus is efficient and robust to the quality of geometry reconstruction, requiring only a few multi-view videos for physical parameter identification. The method is effective in capturing the dynamics of elastic objects, including future prediction and simulation under varying initial configurations. It is also capable of handling heterogeneous materials and complex deformations. The method is compared with existing approaches such as PAC-NeRF and Dynamic 3D Gaussians, showing its superior performance in terms of geometric accuracy and future prediction. Spring-Gaus is also capable of generalizing to new conditions, allowing for dynamic simulation under different environmental conditions. The method is efficient and effective in capturing the dynamics of elastic objects, making it a promising approach for applications in computer vision and robotics.Spring-Gaus is a novel method for reconstructing and simulating elastic objects from multi-view videos. It integrates a 3D Spring-Mass model with 3D Gaussians to capture both the visual appearance and physical dynamics of elastic objects. The method enables accurate reconstruction and simulation of elastic objects, allowing for future prediction and simulation under various initial states and environmental conditions. The core challenge is developing an expressive and efficient physical dynamics model that can capture the complex behaviors of elastic objects, including collisions, deformations, and bouncing. Spring-Gaus uses a learnable system of mass points and springs to model the physical properties of elastic objects. It decouples the appearance and geometry reconstruction from the physical dynamics reconstruction, enabling more effective optimization. The method is evaluated on both synthetic and real-world datasets, demonstrating its ability to accurately reconstruct and simulate elastic objects. Spring-Gaus is efficient and robust to the quality of geometry reconstruction, requiring only a few multi-view videos for physical parameter identification. The method is effective in capturing the dynamics of elastic objects, including future prediction and simulation under varying initial configurations. It is also capable of handling heterogeneous materials and complex deformations. The method is compared with existing approaches such as PAC-NeRF and Dynamic 3D Gaussians, showing its superior performance in terms of geometric accuracy and future prediction. Spring-Gaus is also capable of generalizing to new conditions, allowing for dynamic simulation under different environmental conditions. The method is efficient and effective in capturing the dynamics of elastic objects, making it a promising approach for applications in computer vision and robotics.