ResNet50 outperforms Vision Transformer in efficiency for GI disease classification

A new benchmark study evaluated ResNet50 and Vision Transformer (ViT-B/16) for classifying gastrointestinal diseases from endoscopic images. While both models achieved high accuracy (around 92%) in cross-validation, ResNet50 demonstrated significantly better efficiency with fewer parameters, lower latency, and higher GPU throughput. An edge-oriented transformer, MobileViT-S, was more accurate on a test split but showed greater sensitivity to acquisition artifacts, indicating potential issues with shortcut learning. The research suggests that for resource-constrained environments, an efficient CNN like ResNet50 is a practical choice, though local validation is crucial before deployment. AI

IMPACT Highlights the trade-offs between model accuracy and efficiency, suggesting CNNs remain a viable option for resource-constrained AI applications.

RANK_REASON The cluster contains an academic paper detailing a benchmark study of AI models for a specific classification task. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• GastroVision
• Kvasir dataset
• MobileViT-S
• ResNet50
• ScopeGuide-assisted Colonoscopy Versus Conventional Colonoscopy
• Two one-sided tests procedures in establishing therapeutic equivalence with binary clinical endpoints: fixed sample performances and sample size derformination
• vision transformer

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →