Image resolution changes affect real-time object detection speed
Input Resolution Matters: Real-Time Object Detection Latency
Computer Vision and Pattern RecognitionPerformance
Summary
When computers identify objects in pictures, the time it takes depends on several steps like preparing the image, running the detection, and handling results. The authors found that how big the picture is (its resolution) changes how long these steps take. They made a way to predict total detection time by combining models of each step that change with resolution. By testing on a popular object detector running on a specialized computer, they showed their prediction was more accurate than older methods that ignored resolution effects.
What this means in practice
- •For embedded system engineers: Predict and manage object detection delays in devices with varying input image sizes, improving responsiveness of real-time vision applications.
- •For autonomous vehicle developers: Estimate end-to-end detection latency based on camera resolution to optimize sensor configurations for timely decision making.
Authors
Qingyang Zhang, Fumio Machida, Laura Carnevali
Abstract
We model total latency as the convolution of preprocessing, inference, and postprocessing distributions under a simplifying independence approximation, with selected stage parameters expressed as functions of source-image resolution. Under this assumption, the probability density of the total latency is the convolution of the stage-wise densities, and its cumulative distribution function (CDF) provides the distribution of end-to-end detection time. Each stage is modeled by a parametric distribution (e.g., Exponential, Erlang, Normal, Gamma), with parameters expressed as functions of the source-image resolution. Experiments with YOLOv11n on NVIDIA Jetson Orin NX using COCO2017 images across multiple resolutions assess the proposed models against fixed-parameter baselines using Kolmogorov Smirnov, Anderson Darling, and Cramér von Mises statistics. The results indicate that resolution-aware parameterization can improve distributional approximation in the measured setting, particularly for the more flexible Normal and Gamma models, while the quality of fit remains distribution dependent. Our contribution is a theoretically grounded and lightweight formulation for studying resolution-dependent latency distributions in a measured object detection pipeline.