# Neural Network Generalization: The impact of camera parameters

arXiv: https://arxiv.org/abs/1912.03604 · HTML: https://mlsanity.com/paper/1912.03604/

## Key facts

- Cited by 14 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 8 Dec 2019.

## Built on by (most cited first, 14 of 14)

- [Ethical Considerations for Responsible Data Curation](https://mlsanity.com/paper/2302.03629/) (2023)
- [Neural Camera Simulators](https://mlsanity.com/paper/2104.05237/) (2021)
- [Video-rate multispectral imaging in laparoscopic surgery: First-in-human application](https://mlsanity.com/paper/2105.13901/) (2021)
- [Raw Bayer Pattern Image Synthesis for Computer Vision-oriented Image Signal Processing Pipeline Design](https://mlsanity.com/paper/2110.12823/) (2021)
- [ISETAuto: Detecting vehicles with depth and radiance information](https://mlsanity.com/paper/2101.01843/) (2021)
- [Using simulation to quantify the performance of automotive perception systems](https://mlsanity.com/paper/2303.00983/) (2023)
- [Generalization Across Experimental Parameters in Machine Learning Analysis of High Resolution Transmission Electron Microscopy Datasets](https://mlsanity.com/paper/2306.11853/) (2023)
- [HeteroSwitch: Characterizing and Taming System-Induced Data Heterogeneity in Federated Learning](https://mlsanity.com/paper/2403.04207/) (2024)
- [ISETHDR: A Physics-based Synthetic Radiance Dataset for High Dynamic Range Driving Scenes](https://mlsanity.com/paper/2408.12048/) (2024)
- [Comprehensive Analysis of the Object Detection Pipeline on UAVs](https://mlsanity.com/paper/2203.00306/) (2022)
- [Creating awareness about security and safety on highways to mitigate wildlife-vehicle collisions by detecting and recognizing wildlife fences using deep learning and drone technology](https://mlsanity.com/paper/2301.07174/) (2022)
- [VALERIE22 -- A photorealistic, richly metadata annotated dataset of urban environments](https://mlsanity.com/paper/2308.09632/) (2023)
- [POLAR-Sim: Augmenting NASA's POLAR Dataset for Data-Driven Lunar Perception and Rover Simulation](https://mlsanity.com/paper/2309.12397/) (2023)
- [DiffPhysCam: Differentiable Physics-Based Camera Simulation for Inverse Rendering and Embodied AI](https://mlsanity.com/paper/2508.08831/) (2025)

## Abstract (arXiv)

We quantify the generalization of a convolutional neural network (CNN) trained to identify cars. First, we perform a series of experiments to train the network using one image dataset - either synthetic or from a camera - and then test on a different image dataset. We show that generalization between images obtained with different cameras is roughly the same as generalization between images from a camera and ray-traced multispectral synthetic images. Second, we use ISETAuto, a soft prototyping tool that creates ray-traced multispectral simulations of camera images, to simulate sensor images with a range of pixel sizes, color filters, acquisition and post-acquisition processing. These experiments reveal how variations in specific camera parameters and image processing operations impact CNN generalization. We find that (a) pixel size impacts generalization, (b) demosaicking substantially impacts performance and generalization for shallow (8-bit) bit-depths but not deeper ones (10-bit), and (c) the network performs well using raw (not demosaicked) sensor data for 10-bit pixels.