On-device benchmark and robustness results for lightweight CNNs on MangoLeafDB/MangoLeafDB-C

This Figshare item provides the consolidated, paper-ready artifacts supporting the manuscript “A Comparative Analysis of Lightweight CNNs for Robust and Efficient Mango Leaf Disease Classification”. It includes final figures (PDF), on-device benchmark outputs (JSON) reporting latency statistics (including P90) and throughput for multiple devices (Android smartphones and Raspberry Pi Zero 2 W) using TensorFlow Lite (CPU), robustness results on the MangoLeafDB-C corruption benchmark (JSON/XLSX), clean accuracy summary, and the exported model files (.tflite/.keras) with size summaries. Datasets are not redistributed; MangoLeafDB is available on Kaggle and MangoLeafDB-C can be generated using the provided corruption scripts referenced in the repository.