# Do Image Classifiers Generalize Across Time?

## TL;DR

This paper investigates how well image classifiers trained on static images perform on temporally perturbed images from videos, revealing significant robustness challenges in real-world video environments.

## Contribution

The study introduces two new datasets with human-annotated video-derived images and evaluates the robustness of classifiers and detection models to natural video perturbations.

## Key findings

- Median accuracy drops of 16 and 10 on the new datasets.
- Detection models experience a median mAP drop of 14 points.
- Natural video perturbations significantly impair classifier and detector performance.

## Abstract

We study the robustness of image classifiers to temporal perturbations derived from videos. As part of this study, we construct two datasets, ImageNet-Vid-Robust and YTBB-Robust , containing a total 57,897 images grouped into 3,139 sets of perceptually similar images. Our datasets were derived from ImageNet-Vid and Youtube-BB respectively and thoroughly re-annotated by human experts for image similarity. We evaluate a diverse array of classifiers pre-trained on ImageNet and show a median classification accuracy drop of 16 and 10 on our two datasets. Additionally, we evaluate three detection models and show that natural perturbations induce both classification as well as localization errors, leading to a median drop in detection mAP of 14 points. Our analysis demonstrates that perturbations occurring naturally in videos pose a substantial and realistic challenge to deploying convolutional neural networks in environments that require both reliable and low-latency predictions