DFU Image Classification Cost Reduction

Jan 23, 2025·
Lakshya
· 1 min read
Abstract
This project focuses on Knowledge Distillation & Network Pruning for efficient edge deployment of medical imaging models. By implementing a paired reduction strategy using a ResNet-50 Teacher and MobileNetV2 Student, we achieved a 3.47x inference speedup on CPU and 1.71x on GPU, optimized for edge deployment without sacrificing diagnostic accuracy.
Type
Publication
Pipeline / In Progress
publications

Overview

Diabetic Foot Ulcer (DFU) detection relies on heavyweight models that are often unsuitable for mobile or edge deployment. This project addresses the computational bottleneck by applying Knowledge Distillation (KD) and Network Pruning.

Key Achievements

  1. Architecture: ResNet-50 (Teacher) → MobileNetV2 (Student).
  2. Performance:
    • CPU Inference: 3.47x Speedup.
    • GPU Inference: 1.71x Speedup.
  3. Deployment: Optimized for low-latency healthcare devices.
Computer Vision Edge AI Model Optimization
Authors
Lakshya (he/him)
AI Researcher & Systems Engineer
M.Tech AI & ML student specializing in Multimodal Representation Learning and Advanced Systems Engineering.