Practical GPU Choices for Earth Observation: ResNet-50 Training Throughput on Integrated, Laptop, and Cloud Accelerators
This work addresses the challenge of scalable geospatial analytics for researchers and practitioners by demonstrating the feasibility of using consumer and free cloud GPUs, though it is incremental as it applies an existing method to new hardware configurations.
This paper tackled the problem of training deep learning models for land use and land cover classification on Earth observation data by benchmarking ResNet-50 training across different GPUs, achieving up to a 2x speed-up on NVIDIA RTX 3060 and Tesla T4 compared to an Apple M3 Pro baseline while maintaining high accuracy on the EuroSAT dataset.
This project implements a ResNet-based pipeline for land use and land cover (LULC) classification on Sentinel-2 imagery, benchmarked across three heterogeneous GPUs. The workflow automates data acquisition, geospatial preprocessing, tiling, model training, and visualization, and is fully containerized for reproducibility. Performance evaluation reveals up to a 2x training speed-up on an NVIDIA RTX 3060 and a Tesla T4 compared to the Apple M3 Pro baseline, while maintaining high classification accuracy on the EuroSAT dataset. These results demonstrate the feasibility of deploying deep learning LULC models on consumer and free cloud GPUs for scalable geospatial analytics.