# Neural Re-Simulation for Generating Bounces in Single Images

## TL;DR

This paper presents a neural re-simulation method that refines physically simulated trajectories of virtual objects in still images to produce visually plausible videos of interactions and collisions.

## Contribution

It introduces neural re-simulation, a novel approach to correct and enhance traditional physics-based simulations for realistic video generation from single images.

## Key findings

- Consistent improvement over baseline methods in synthetic and real images.
- Effective generation of plausible object interactions and bounces.
- Trained on 50k synthetic scenes with physical simulations.

## Abstract

We introduce a method to generate videos of dynamic virtual objects plausibly interacting via collisions with a still image's environment. Given a starting trajectory, physically simulated with the estimated geometry of a single, static input image, we learn to 'correct' this trajectory to a visually plausible one via a neural network. The neural network can then be seen as learning to 'correct' traditional simulation output, generated with incomplete and imprecise world information, to obtain context-specific, visually plausible re-simulated output, a process we call neural re-simulation. We train our system on a set of 50k synthetic scenes where a virtual moving object (ball) has been physically simulated. We demonstrate our approach on both our synthetic dataset and a collection of real-life images depicting everyday scenes, obtaining consistent improvement over baseline alternatives throughout.