# Floors are Flat: Leveraging Semantics for Real-Time Surface Normal Prediction

## TL;DR

This paper introduces four practical insights that significantly enhance real-time surface normal and semantic label prediction from a single RGB image, achieving improved accuracy and efficiency on mobile devices.

## Contribution

The paper presents four novel, simple techniques for improving deep learning models for surface normal prediction, including data denoising, mixed data training, joint prediction, and model simplification.

## Key findings

- Improved accuracy on multiple datasets.
- Real-time performance at 12 fps on mobile devices.
- Effective use of combined real and synthetic data.

## Abstract

We propose 4 insights that help to significantly improve the performance of deep learning models that predict surface normals and semantic labels from a single RGB image. These insights are: (1) denoise the "ground truth" surface normals in the training set to ensure consistency with the semantic labels; (2) concurrently train on a mix of real and synthetic data, instead of pretraining on synthetic and finetuning on real; (3) jointly predict normals and semantics using a shared model, but only backpropagate errors on pixels that have valid training labels; (4) slim down the model and use grayscale instead of color inputs. Despite the simplicity of these steps, we demonstrate consistently improved results on several datasets, using a model that runs at 12 fps on a standard mobile phone.