diff --git a/ppcls/configs/reid/MetaBIN_ResNet50_single_source.yaml b/ppcls/configs/reid/MetaBIN_ResNet50_single_source.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..a8808a7a090096f121d916fd2e94713b24dec22b
--- /dev/null
+++ b/ppcls/configs/reid/MetaBIN_ResNet50_single_source.yaml
@@ -0,0 +1,282 @@
+# global configs
+Global:
+  checkpoints: null
+  pretrained_model: null #"metabin_resnet50_final"
+  output_dir: "./output/"
+  device: "gpu"
+  iter_per_epoch: 145
+  warmup_iter: 10
+  save_interval: 40
+  eval_during_train: True
+  eval_interval: 10
+  epochs: &epochs 120
+  print_batch_step: 20
+  use_visualdl: False
+  eval_mode: "retrieval"
+  retrieval_feature_from: "features" # 'backbone' or 'features'
+  re_ranking: False
+  # used for static mode and model export
+  image_shape: [3, 256, 128]
+  save_inference_dir: "./inference"
+  train_mode: 'metabin'
+
+# model architecture
+Arch:
+  name: "RecModel"
+  Backbone:
+    name: "ResNet50_metabin"
+    pretrained: False # "metabin_resnet50_backbone_pretrained"
+    bias_lr_factor: 2.0
+    gate_lr_factor: 20.0
+  BackboneStopLayer:
+    name: "flatten"
+  Neck:
+    name: BNNeck
+    num_features: &feat_dim 2048
+    weight_attr:
+      initializer:
+        name: Constant
+        value: 1.0
+  Head:
+    name: "FC"
+    embedding_size: *feat_dim
+    class_num: &class_num 751
+    weight_attr:
+      initializer:
+        name: KaimingUniform
+        negative_slope: 2.23606   # math.sqrt(5)
+        nonlinearity: "leaky_relu"
+    bias_attr: False
+
+# data loader for train and eval
+DataLoader:
+  Train:
+    dataset:
+      name: "Market1501"
+      image_root: "./dataset/"
+      cls_label_path: "bounding_box_train"
+      backend: "pil"
+      transform_ops:
+        - ResizeImage:
+            size: [128, 256]
+            return_numpy: False
+            interpolation: "bicubic"
+            backend: "pil"
+        - RandFlipImage:
+            flip_code: 1
+        - Pad:
+            padding: 10
+        - RandCropImageV2:
+            size: [128, 256]
+        - ColorJitter:
+            brightness: 0.15
+            contrast: 0.15
+            saturation: 0.1
+            hue: 0.1
+        - ToTensor:
+        - Normalize:
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+    sampler:
+      name: NaiveIdentityBatchSampler
+      batch_size: 96
+      num_instances: 4
+      drop_last: True
+    loader:
+      num_workers: 4
+      use_shared_memory: True
+      
+  Metalearning:
+    Train:
+      dataset:
+        name: "Market1501"
+        image_root: "./dataset/"
+        cls_label_path: "bounding_box_train"
+        backend: "pil"
+        transform_ops:
+          - ResizeImage:
+              size: [128, 256]
+              return_numpy: False
+              interpolation: "bicubic"
+              backend: "pil"
+          - RandFlipImage:
+              flip_code: 1
+          - Pad:
+              padding: 10
+          - RandCropImageV2:
+              size: [128, 256]
+          - ColorJitter:
+              brightness: 0.15
+              contrast: 0.15
+              saturation: 0.1
+              hue: 0.1
+          - ToTensor:
+          - Normalize:
+              mean: [0.485, 0.456, 0.406]
+              std: [0.229, 0.224, 0.225]
+      sampler:
+        name: DomainShuffleBatchSampler
+        batch_size: 96
+        num_instances: 4
+        drop_last: True
+        camera_to_domain: True
+      loader:
+        num_workers: 4
+        use_shared_memory: True
+
+  Eval:
+    Query:
+      dataset:
+        name: "DukeMTMC"
+        image_root: "./dataset/"
+        cls_label_path: "query"
+        backend: "pil"
+        transform_ops:
+          - ResizeImage:
+              size: [128, 256]
+              return_numpy: False
+              interpolation: "bicubic"
+              backend: "pil"
+          - ToTensor:
+          - Normalize:
+              mean: [0.485, 0.456, 0.406]
+              std: [0.229, 0.224, 0.225]
+      sampler:
+        name: DistributedBatchSampler
+        batch_size: 128
+        drop_last: False
+        shuffle: False
+      loader:
+        num_workers: 4
+        use_shared_memory: True
+
+    Gallery:
+      dataset:
+        name: "DukeMTMC"
+        image_root: "./dataset/"
+        cls_label_path: "bounding_box_test"
+        backend: "pil"
+        transform_ops:
+          - ResizeImage:
+              size: [128, 256]
+              return_numpy: False
+              interpolation: "bicubic"
+              backend: "pil"
+          - ToTensor:
+          - Normalize:
+              mean: [0.485, 0.456, 0.406]
+              std: [0.229, 0.224, 0.225]
+      sampler:
+        name: DistributedBatchSampler
+        batch_size: 128
+        drop_last: False
+        shuffle: False
+      loader:
+        num_workers: 4
+        use_shared_memory: True
+
+# loss function config for traing/eval process
+Loss:
+  Train:
+    - CELossForMetaBIN:
+        weight: 1.0
+        epsilon: 0.1
+    - TripletLossForMetaBIN:
+        weight: 1.0
+        margin: 0.3
+        feature_from: "backbone"
+    - IntraDomainScatterLoss:
+        weight: 1.0
+        normalize_feature: True
+        feature_from: "backbone"
+    - InterDomainShuffleLoss:
+        weight: 1.0
+        normalize_feature: False
+        feature_from: "backbone"
+  Basic:
+    - CELossForMetaBIN:
+        weight: 1.0
+        epsilon: 0.1
+    - TripletLossForMetaBIN:
+        weight: 1.0
+        margin: 0.3
+        feature_from: "backbone"
+  MetaTrain:
+    - CELossForMetaBIN:
+        weight: 1.0
+        epsilon: 0.1
+    - TripletLossForMetaBIN:
+        weight: 1.0
+        margin: 0.3
+        feature_from: "backbone"
+    - IntraDomainScatterLoss:
+        weight: 1.0
+        normalize_feature: True
+        feature_from: "backbone"
+    - InterDomainShuffleLoss:
+        weight: 1.0
+        normalize_feature: False
+        feature_from: "backbone"
+  MetaTest:
+    - CELossForMetaBIN:
+        weight: 1.0
+        epsilon: 0.1
+    - TripletLossForMetaBIN:
+        weight: 1.0
+        margin: 0.3
+        feature_from: "backbone"
+  Eval:
+    - TripletLossForMetaBIN:
+        weight: 1.0
+        margin: 0.3
+        feature_from: "backbone"
+
+Optimizer:
+  - Momentum:
+      scope: ".*(conv|batch_norm|instance_norm|feat_bn|fc)"
+      lr:
+        name: MultiStepDecay
+        epochs: *epochs
+        learning_rate: 0.01
+        step_each_epoch: 145
+        milestones: [50, 90]
+        gamma: 0.1
+        warmup_epoch: 10
+        warmup_start_lr: 0.0001
+        by_epoch: False
+        last_epoch: -1
+      momentum: 0.9
+      regularizer:
+        name: "L2"
+        coeff: 0.0005
+  - Momentum:
+      scope: "backbone.*gate"
+      lr:
+        name: Constant
+        learning_rate: 0.01
+        last_epoch: 0
+      momentum: 0.9
+  - SGD:
+      scope: "RecModel"
+      lr:
+        name: Cyclic
+        epochs: *epochs
+        step_each_epoch: 145
+        base_learning_rate: 0.001
+        max_learning_rate: 0.1
+        warmup_epoch: 0
+        warmup_start_lr: 1
+        step_size_up: 1095
+        step_size_down: 1095
+        by_epoch: False
+        last_epoch: 0
+
+AMP:
+  scale_loss: 65536
+  use_dynamic_loss_scaling: True
+
+Metric:
+  Eval:
+    - Recallk:
+        topk: [1, 5, 10]
+    - mAP: {}
\ No newline at end of file
diff --git a/ppcls/engine/train/__init__.py b/ppcls/engine/train/__init__.py
index a3fea6938905cda3a1b65d1fe82321d025a7d6b3..50bf9037f4982354724d56f5814f47cf8b92decc 100644
--- a/ppcls/engine/train/__init__.py
+++ b/ppcls/engine/train/__init__.py
@@ -15,3 +15,4 @@ from ppcls.engine.train.train import train_epoch
 from ppcls.engine.train.train_fixmatch import train_epoch_fixmatch
 from ppcls.engine.train.train_fixmatch_ccssl import train_epoch_fixmatch_ccssl
 from ppcls.engine.train.train_progressive import train_epoch_progressive
+from ppcls.engine.train.train_metabin import train_epoch_metabin
diff --git a/ppcls/engine/train/train_metabin.py b/ppcls/engine/train/train_metabin.py
new file mode 100644
index 0000000000000000000000000000000000000000..d9348408114412c93e5ec37b61c329c902f20d57
--- /dev/null
+++ b/ppcls/engine/train/train_metabin.py
@@ -0,0 +1,251 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# reference: https://arxiv.org/abs/2011.14670v2
+
+from __future__ import absolute_import, division, print_function
+
+import time
+import paddle
+import numpy as np
+from collections import defaultdict
+
+from ppcls.engine.train.utils import update_loss, update_metric, log_info, type_name
+from ppcls.utils import profiler
+from ppcls.data import build_dataloader
+from ppcls.arch.backbone.variant_models.resnet_variant import MetaBIN, BINGate
+from ppcls.loss import build_loss
+
+
+def train_epoch_metabin(engine, epoch_id, print_batch_step):
+    tic = time.time()
+
+    if not hasattr(engine, "train_dataloader_iter"):
+        engine.train_dataloader_iter = iter(engine.train_dataloader)
+
+    if not hasattr(engine, "meta_dataloader"):
+        engine.meta_dataloader = build_dataloader(
+            config=engine.config['DataLoader']['Metalearning'],
+            mode='Train',
+            device=engine.device)
+        engine.meta_dataloader_iter = iter(engine.meta_dataloader)
+
+    num_domain = engine.train_dataloader.dataset.num_cams
+    for iter_id in range(engine.iter_per_epoch):
+        # fetch data batch from dataloader
+        try:
+            train_batch = next(engine.train_dataloader_iter)
+        except Exception:
+            engine.train_dataloader_iter = iter(engine.train_dataloader)
+            train_batch = next(engine.train_dataloader_iter)
+
+        try:
+            mtrain_batch, mtest_batch = get_meta_data(
+                engine.meta_dataloader_iter, num_domain)
+        except Exception:
+            engine.meta_dataloader_iter = iter(engine.meta_dataloader)
+            mtrain_batch, mtest_batch = get_meta_data(
+                engine.meta_dataloader_iter, num_domain)
+
+        profiler.add_profiler_step(engine.config["profiler_options"])
+        if iter_id == 5:
+            for key in engine.time_info:
+                engine.time_info[key].reset()
+        engine.time_info["reader_cost"].update(time.time() - tic)
+
+        train_batch_size = train_batch[0].shape[0]
+        mtrain_batch_size = mtrain_batch[0].shape[0]
+        mtest_batch_size = mtest_batch[0].shape[0]
+        if not engine.config["Global"].get("use_multilabel", False):
+            train_batch[1] = train_batch[1].reshape([train_batch_size, -1])
+            mtrain_batch[1] = mtrain_batch[1].reshape([mtrain_batch_size, -1])
+            mtest_batch[1] = mtest_batch[1].reshape([mtest_batch_size, -1])
+
+        engine.global_step += 1
+
+        if engine.global_step == 1:  # update model (without gate) to warmup
+            for i in range(engine.config["Global"]["warmup_iter"] - 1):
+                out, basic_loss_dict = basic_update(engine, train_batch)
+                loss_dict = basic_loss_dict
+                try:
+                    train_batch = next(engine.train_dataloader_iter)
+                except Exception:
+                    engine.train_dataloader_iter = iter(
+                        engine.train_dataloader)
+                    train_batch = next(engine.train_dataloader_iter)
+
+        out, basic_loss_dict = basic_update(engine=engine, batch=train_batch)
+        mtrain_loss_dict, mtest_loss_dict = metalearning_update(
+            engine=engine, mtrain_batch=mtrain_batch, mtest_batch=mtest_batch)
+        loss_dict = {
+            **
+            {"train_" + key: value
+             for key, value in basic_loss_dict.items()}, ** {
+                 "mtrain_" + key: value
+                 for key, value in mtrain_loss_dict.items()
+             }, **
+            {"mtest_" + key: value
+             for key, value in mtest_loss_dict.items()}
+        }
+        # step lr (by iter)
+        # the last lr_sch is cyclic_lr
+        for i in range(len(engine.lr_sch) - 1):
+            if not getattr(engine.lr_sch[i], "by_epoch", False):
+                engine.lr_sch[i].step()
+        # update ema
+        if engine.ema:
+            engine.model_ema.update(engine.model)
+
+        # below code just for logging
+        # update metric_for_logger
+        update_metric(engine, out, train_batch, train_batch_size)
+        # update_loss_for_logger
+        update_loss(engine, loss_dict, train_batch_size)
+        engine.time_info["batch_cost"].update(time.time() - tic)
+        if iter_id % print_batch_step == 0:
+            log_info(engine, train_batch_size, epoch_id, iter_id)
+        tic = time.time()
+
+    # step lr(by epoch)
+    # the last lr_sch is cyclic_lr
+    for i in range(len(engine.lr_sch) - 1):
+        if getattr(engine.lr_sch[i], "by_epoch", False) and \
+                type_name(engine.lr_sch[i]) != "ReduceOnPlateau":
+            engine.lr_sch[i].step()
+
+
+def setup_opt(engine, stage):
+    assert stage in ["train", "mtrain", "mtest"]
+    opt = defaultdict()
+    if stage == "train":
+        opt["bn_mode"] = "general"
+        opt["enable_inside_update"] = False
+        opt["lr_gate"] = 0.0
+    elif stage == "mtrain":
+        opt["bn_mode"] = "hold"
+        opt["enable_inside_update"] = False
+        opt["lr_gate"] = 0.0
+    elif stage == "mtest":
+        norm_lr = engine.lr_sch[1].last_lr
+        cyclic_lr = engine.lr_sch[2].get_lr()
+        engine.lr_sch[2].step()  # update cyclic learning rate
+        opt["bn_mode"] = "hold"
+        opt["enable_inside_update"] = True
+        opt["lr_gate"] = norm_lr * cyclic_lr
+    for layer in engine.model.sublayers():
+        if isinstance(layer, MetaBIN):
+            layer.setup_opt(opt)
+
+
+def reset_opt(model):
+    for layer in model.sublayers():
+        if isinstance(layer, MetaBIN):
+            layer.reset_opt()
+
+
+def get_meta_data(meta_dataloader_iter, num_domain):
+    """
+    fetch data batch from dataloader then divide the batch by domains
+    """
+    list_all = np.random.permutation(num_domain)
+    list_mtrain = list(list_all[:num_domain // 2])
+    batch = next(meta_dataloader_iter)
+    domain_idx = batch[2]
+    cnt = 0
+    for sample in list_mtrain:
+        if cnt == 0:
+            is_mtrain_domain = domain_idx == sample
+        else:
+            is_mtrain_domain = paddle.logical_or(is_mtrain_domain,
+                                                 domain_idx == sample)
+        cnt += 1
+
+    # mtrain_batch
+    if not any(is_mtrain_domain):
+        mtrain_batch = None
+        raise RuntimeError
+    else:
+        mtrain_batch = dict()
+        mtrain_batch = [batch[i][is_mtrain_domain] for i in range(len(batch))]
+
+    # mtest_batch
+    is_mtest_domains = is_mtrain_domain == False
+    if not any(is_mtest_domains):
+        mtest_batch = None
+        raise RuntimeError
+    else:
+        mtest_batch = dict()
+        mtest_batch = [batch[i][is_mtest_domains] for i in range(len(batch))]
+    return mtrain_batch, mtest_batch
+
+
+def forward(engine, batch, loss_func):
+    batch_info = defaultdict()
+    batch_info = {"label": batch[1], "domain": batch[2]}
+    amp_level = engine.config["AMP"].get("level", "O1").upper()
+    with paddle.amp.auto_cast(
+            custom_black_list={"flatten_contiguous_range", "greater_than"},
+            level=amp_level):
+        out = engine.model(batch[0], batch[1])
+        loss_dict = loss_func(out, batch_info)
+    return out, loss_dict
+
+
+def backward(engine, loss, optimizer):
+    scaled = engine.scaler.scale(loss)
+    scaled.backward()
+    engine.scaler.minimize(optimizer, scaled)
+    for layer in engine.model.sublayers():
+        if isinstance(layer, BINGate):
+            layer.clip_gate()
+
+
+def basic_update(engine, batch):
+    setup_opt(engine, "train")
+    train_loss_func = build_loss(engine.config["Loss"]["Basic"])
+    out, train_loss_dict = forward(engine, batch, train_loss_func)
+    train_loss = train_loss_dict["loss"]
+    backward(engine, train_loss, engine.optimizer[0])
+    engine.optimizer[0].clear_grad()
+    reset_opt(engine.model)
+    return out, train_loss_dict
+
+
+def metalearning_update(engine, mtrain_batch, mtest_batch):
+    # meta train
+    mtrain_loss_func = build_loss(engine.config["Loss"]["MetaTrain"])
+    setup_opt(engine, "mtrain")
+
+    mtrain_batch_info = defaultdict()
+    mtrain_batch_info = {"label": mtrain_batch[1], "domain": mtrain_batch[2]}
+    out = engine.model(mtrain_batch[0], mtrain_batch[1])
+    mtrain_loss_dict = mtrain_loss_func(out, mtrain_batch_info)
+    mtrain_loss = mtrain_loss_dict["loss"]
+    engine.optimizer[1].clear_grad()
+    mtrain_loss.backward()
+
+    # meta test
+    mtest_loss_func = build_loss(engine.config["Loss"]["MetaTest"])
+    setup_opt(engine, "mtest")
+
+    out, mtest_loss_dict = forward(engine, mtest_batch, mtest_loss_func)
+    engine.optimizer[1].clear_grad()
+    mtest_loss = mtest_loss_dict["loss"]
+    backward(engine, mtest_loss, engine.optimizer[1])
+
+    engine.optimizer[0].clear_grad()
+    engine.optimizer[1].clear_grad()
+    reset_opt(engine.model)
+
+    return mtrain_loss_dict, mtest_loss_dict