Face Recognition Fine‑Tuning — Quick Start

1. Setup Python 3.8+ and PyTorch (CUDA optional but recommended). Install deps (minimal): numpy, tqdm, torch, torchvision.

2. Data layout

Place images under class-labeled folders:

  <class_id_1>/ 
    img1.jpg
    img2.png
    ...
  <class_id_2>/ 
    img3.jpg
    img4.png
    ...
  <class_id_3>/ 
    ...

3. Start fine-tuning with four category images

• --model_name: mobilefacenet | ir_se
• --classifier_type: arcface | cosface | FC
• --aug_type: standard | strong | none
• --phase: head_only | last_block
• --train_batch_size / --test_batch_size
• --max_epoch, --optimizer (adamw | SGD)
• --save_plot → save loss/acc plots

MobileFaceNet + ArcFace (last block fine‑tune):

python .\fine_tune_main.py --root_dir .\data --model_name mobilefacenet --classifier_type arcface --phase last_block --optimizer adamw --save_plot

MobileFaceNet + CosFace

python .\fine_tune_main.py --root_dir .\data --model_name mobilefacenet --classifier_type cosface --phase last_block --optimizer adamw --save_plot

4. Testing python test_model.py --classifier_type combined