Code and Dataset Manifests for "Benchmarking Face Recognition Architectures Across Server and Edge Platforms"
Year: 2026
Type: book
Abstract: This repository contains the full code and reproducibility materials for the study "Benchmarking Face Recognition Architectures Across Server and Edge Platforms." The repository includes:- Dataset acquisition and verification scripts for UMDFaces and VGGFace2 (Kaggle mirrors), including a SHA-256 checksum manifest for the exact archives used.- Deterministic subsampling code and the resulting identity/image manifests (40 identities x 30 images for UMDFaces; 20 identities × <= 50 images for VGGFace2).- Embedding-generation scripts for all eight evaluated architectures (ArcFace, DeepID, FaceNet, SFace, GhostFaceNet, OpenFace, MobileFaceNet, ResNet34), implemented via DeepFace and InsightFace.- Evaluation code for both verification (many-to-many cosine similarity) and identification (gallery-based ranking) protocols.- Latency, throughput, and RAM/CPU/GPU resource-usage measurement scripts used to produce tables and figures.- Robustness-analysis script.- A pinned Conda environment specification and a step-by-step README for reproducing all reported results on equivalent hardware. All experiments were conducted using publicly available pretrained model weights only; no model was fine-tuned on UMDFaces or VGGFace2.
Institution: Astana IT University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7204238332