FACTORS THAT AFFECT IMAGE RECOGNITION EFFICIENCY WITH HUGGING FACE MODELS

Abstract

This study investigated the factors influencing the performance of image recognition systems, with particular attention to image quality, image type, and computational hardware. The objective was to identify optimal configurations for real-world applications by examining the trade-offs among accuracy, processing time, and computational efficiency. Pre-trained image classification models were evaluated on datasets comprising high-, low-, and mixed-resolution images, including natural landscapes, medical scans, and complex scenes. Each experimental condition was assessed in both CPU and GPU environments to measure recognition accuracy and processing time. The results indicated that high-resolution images generally improved recognition accuracy but also increased processing time and computational demands. Image complexity was found to affect both accuracy and inference time, particularly for deeper models. GPUs consistently outperformed CPUs, reducing inference time by more than 72 percent in some cases, especially when processing high-resolution or complex images. Nevertheless, CPUs provided acceptable performance for simpler tasks or in hardware-constrained environments. Overall, these findings provide practical guidance for selecting appropriate hardware and image configurations to optimize the performance of image recognition systems.