[2603.02142] Is Bigger Always Better? Efficiency Analysis in Resource-Constrained Small Object Detection

Computer Science > Computer Vision and Pattern Recognition arXiv:2603.02142 (cs) [Submitted on 2 Mar 2026] Title:Is Bigger Always Better? Efficiency Analysis in Resource-Constrained Small Object Detection Authors:Kwame Mbobda-Kuate, Gabriel Kasmi View a PDF of the paper titled Is Bigger Always Better? Efficiency Analysis in Resource-Constrained Small Object Detection, by Kwame Mbobda-Kuate and 1 other authors View PDF HTML (experimental) Abstract:Scaling laws assume larger models trained on more data consistently outperform smaller ones -- an assumption that drives model selection in computer vision but remains untested in resource-constrained Earth observation (EO). We conduct a systematic efficiency analysis across three scaling dimensions: model size, dataset size, and input resolution, on rooftop PV detection in Madagascar. Optimizing for model efficiency (mAP$_{50}$ per unit of model size), we find a consistent efficiency inversion: YOLO11N achieves both the highest efficiency ($24\times$ higher than YOLO11X) and the highest absolute mAP$_{50}$ (0.617). Resolution is the dominant resource allocation lever ($+$120% efficiency gain), while additional data yields negligible returns at low resolution. These findings are robust to the deployment objective: small high-resolution configurations are Pareto-dominant across all 44 setups in the joint accuracy-throughput space, leaving no tradeoff to resolve. In data-scarce EO, bigger is not just unnecessary: it can be worse. Co...