PaddleX/paddlex/repo_apis/PaddleClas_api/configs/ResNet50_face.yaml

# global configs
Global:
  checkpoints: null
  pretrained_model: null
  output_dir: ./output/
  device: gpu
  save_interval: 10
  eval_during_train: True
  eval_interval: 1
  epochs: 25
  print_batch_step: 20
  use_visualdl: False
  eval_mode: face_recognition
  retrieval_feature_from: backbone 
  flip_test: True
  feature_normalize: False
  re_ranking: False
  use_dali: False
  # used for static mode and model export
  image_shape: [3, 112, 112]
  save_inference_dir: ./inference

AMP:
  use_amp: True
  use_fp16_test: False
  scale_loss: 27648
  use_dynamic_loss_scaling: True
  use_promote: False
  # O1: mixed fp16
  level: O1

# model architecture
Arch:
  name: RecModel
  infer_output_key: features
  infer_add_softmax: False

  Backbone:
    name: ResNet50
    max_pool: False
    stride_list: [1, 2, 2, 2, 2]
    class_num: 512
  Head:
    name: ArcMargin
    embedding_size: 512
    class_num: 995
    margin: 0.5
    scale: 64
# loss function config for traing/eval process
Loss:
  Train:
    - CELoss:
        weight: 1.0

Optimizer:
  name: AdamW
  beta1: 0.9
  beta2: 0.999
  epsilon: 1e-8
  weight_decay: 0.05
  one_dim_param_no_weight_decay: True
  lr:
    # for 8 cards
    name: Cosine
    learning_rate: 4e-3  # lr 4e-3 for total_batch_size 1024
    eta_min: 1e-6
    warmup_epoch: 1
    warmup_start_lr: 0

# data loader for train and eval
DataLoader:
  Train:
    dataset:
      name: ImageNetDataset
      image_root: dataset/MS1M_v3/
      cls_label_path: dataset/MS1M_v3/label.txt
      transform_ops:
        - DecodeImage:
            to_rgb: True
            channel_first: False
            backend: cv2
        - RandFlipImage:
            flip_code: 1
        - ResizeImage:
            size: [112, 112]
            return_numpy: False
            interpolation: bilinear
            backend: cv2
        - NormalizeImage:
            scale: 1.0/255.0
            mean: [0.5, 0.5, 0.5]
            std: [0.5, 0.5, 0.5]
            order: hwc
    sampler:
      name: DistributedBatchSampler
      batch_size: 128
      drop_last: False
      shuffle: True
    loader:
      num_workers: 8
      use_shared_memory: True

  Eval:
    dataset:
      name: FaceEvalDataset
      transform_ops:
        - DecodeImage:
            to_rgb: True
            channel_first: False
            backend: cv2
        - ResizeImage:
            size: [112, 112]
            return_numpy: False
            interpolation: bilinear
            backend: cv2
        - NormalizeImage:
            scale: 1.0/255.0
            mean: [0.5, 0.5, 0.5]
            std: [0.5, 0.5, 0.5]
            order: hwc
    sampler:
      name: DistributedBatchSampler
      batch_size: 32
      drop_last: False
      shuffle: False
    loader:
      num_workers: 4
      use_shared_memory: True

Metric:
  Eval:
    - FaceAccuracy: {}