Made figure clearer

Author: eikehmueller

README.md

@@ -21,7 +21,10 @@ End-to-end experiments evaluate the impact of conservation constraints on model
 
 In contrast to a purely data-driven Lagrangian neural network (bottom), our new method (top) is stable and matches the exact solution. The solution is physical in the sense that the trajectory lies in a fixed plane.
 
+#### Our method
 ![Our method (Lagrangian NN with built-in conservation laws)](figures/trajectories_schwarzschild_rot.png)
+
+#### Reference method
 ![Reference method (data-driven Lagrangian NN)](figures/trajectories_schwarzschild.png)
 
 The figures (taken from [our paper](https://arxiv.org/abs/2209.11661)) shows the simulation of a particle trajectory in a radially symmetric Schwarzschild metric.