The figure below shows some views from the cluttered desk area belonging to the ‘Apartment’ scene presented in the EyeFul Tower Dataset. We notice a clear visual improvement in the reconstruction fidelity when we use NeRF supervision to train a Scaffold-GS instance.
![](https://mscvprojects.ri.cmu.edu/2024team2/wp-content/uploads/sites/100/2024/05/Screenshot-2024-05-12-at-6.06.04 PM-1024x334.png)
![](https://mscvprojects.ri.cmu.edu/2024team2/wp-content/uploads/sites/100/2024/05/Screenshot-2024-05-12-at-6.06.13 PM-1024x332.png)
![](https://mscvprojects.ri.cmu.edu/2024team2/wp-content/uploads/sites/100/2024/05/Screenshot-2024-05-12-at-6.06.19 PM-1024x334.png)
![](https://mscvprojects.ri.cmu.edu/2024team2/wp-content/uploads/sites/100/2024/05/Screenshot-2024-05-12-at-6.06.25 PM-1024x376.png)
Figure: Improvement on Scaffold-GS using NeRF supervision