Merge pull request PaddlePaddle#119

add FaceViT

plsc/configs/FaceRecognition/FaceViT_tiny_patch9_112_WebFace42M_ArcFace_0.3_droppath005_mask0_1n8c_dp_mp_fp16o2.yaml

@@ -0,0 +1,132 @@
+# global configs
+Global:
+  task_type: recognition
+  train_epoch_func: defualt_train_one_epoch
+  eval_func: face_verification_eval
+  checkpoint: null
+  pretrained_model: null
+  output_dir: ./output/
+  device: gpu
+  save_interval: 1
+  max_num_latest_checkpoint: 0
+  eval_during_train: True
+  eval_interval: 4000
+  eval_unit: "step"
+  accum_steps: 1
+  epochs: 40
+  print_batch_step: 200
+  use_visualdl: True
+  seed: 2022
+
+# FP16 setting
+FP16:
+  level: O2
+  GradScaler:
+    init_loss_scaling: 27648.0
+    # no_unscale_list: ['dist']
+
+DistributedStrategy:
+  data_parallel: True
+
+# model architecture
+Model:
+  name: FaceViT_tiny_patch9_112
+  num_features : 512
+  class_num: 2059906
+  pfc_config:
+    sample_ratio: 0.3
+    model_parallel: True
+    sparse_update: False
+
+# loss function config for traing/eval process
+Loss:
+  Train:
+    - MarginLoss:
+        m1: 1.0
+        m2: 0.0
+        m3: 0.4
+        s: 64.0
+        model_parallel: True
+        weight: 1.0
+
+LRScheduler:
+  name: Poly
+  learning_rate: 0.001
+  decay_unit: step
+  warmup_epochs: 4
+  warmup_end_lr: 0.001
+
+Optimizer:
+  name: AdamW
+  betas: (0.9, 0.999)
+  epsilon: 1e-8
+  weight_decay: 0.1
+  use_master_param: False
+  exp_avg_force_fp32: True
+  no_weight_decay_name: ["pos_embed", "norm"]
+  grad_clip:
+    name: ClipGradByGlobalNorm
+    clip_norm: 5.0
+    no_clip_list: ['dist']
+
+
+# data loader for train and eval
+DataLoader:
+  Train:
+    dataset:
+      name: FaceIdentificationDataset
+      image_root: ./dataset/WebFace42M
+      cls_label_path: ./dataset/WebFace42M/label.txt
+      transform_ops:
+        - DecodeImage:
+            to_rgb: True
+            channel_first: False
+        - RandFlipImage:
+            flip_code: 1
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.5, 0.5, 0.5]
+            std: [0.5, 0.5, 0.5]
+            order: ''
+        - ToCHWImage: 
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 512
+      drop_last: False
+      shuffle: True
+    loader:
+      num_workers: 8
+      use_shared_memory: True
+
+  Eval:
+    dataset: 
+      name: FaceVerificationDataset
+      image_root: ./dataset/WebFace42M/agedb_30
+      cls_label_path: ./dataset/WebFace42M/agedb_30/label.txt
+      transform_ops:
+        - DecodeImage:
+            to_rgb: True
+            channel_first: False
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.5, 0.5, 0.5]
+            std: [0.5, 0.5, 0.5]
+            order: ''
+        - ToCHWImage:
+    sampler:
+      name: BatchSampler
+      batch_size: 128
+      drop_last: False
+      shuffle: False
+    loader:
+      num_workers: 0
+      use_shared_memory: True
+
+Metric:
+  Eval:
+    - LFWAcc:
+        flip_test: True
+
+Export:
+  export_type: onnx
+  input_shape: [None, 3, 112, 112]

plsc/models/__init__.py

@@ -19,6 +19,7 @@
 from .vision_transformer import *
 from .deit import *
 from .iresnet import *
+from .face_vit import *
 
 __all__ = ["build_model"]