Merge pull request #656 from flyseaworld/develop

add multilabel feature based develop

README.md

@@ -82,6 +82,7 @@ PaddleClas is a toolset for image classification tasks prepared for the industry
 - Advanced tutorials
     - [Knowledge distillation](./docs/en/advanced_tutorials/distillation/distillation_en.md)
     - [Data augmentation](./docs/en/advanced_tutorials/image_augmentation/ImageAugment_en.md)
+    - [Multilabel classification](./docs/en/advanced_tutorials/multilabel/multilabel_en.md)
 - Applications
     - [Transfer learning](./docs/en/application/transfer_learning_en.md)
     - [Pretrained model with 100,000 categories](./docs/en/application/transfer_learning_en.md)

README_cn.md

@@ -83,6 +83,7 @@
 - 高阶使用
     - [知识蒸馏](./docs/zh_CN/advanced_tutorials/distillation/distillation.md)
     - [数据增广](./docs/zh_CN/advanced_tutorials/image_augmentation/ImageAugment.md)
+    - [多标签分类](./docs/zh_CN/advanced_tutorials/multilabel/multilabel.md)
 - 特色拓展应用
     - [迁移学习](./docs/zh_CN/application/transfer_learning.md)
     - [10万类图像分类预训练模型](./docs/zh_CN/application/transfer_learning.md)

configs/quick_start/ResNet50_vd_multilabel.yaml

+mode: 'train'
+ARCHITECTURE:
+    name: 'ResNet50_vd'
+
+pretrained_model: "./pretrained/ResNet50_vd_pretrained"
+model_save_dir: "./output/"
+classes_num: 33
+total_images: 17463
+save_interval: 1
+validate: True
+valid_interval: 1
+epochs: 10
+topk: 1
+image_shape: [3, 224, 224]
+
+multilabel: True
+
+use_mix: False
+ls_epsilon: 0.1
+
+LEARNING_RATE:
+    function: 'Cosine'          
+    params:                   
+        lr: 0.07           
+
+OPTIMIZER:
+    function: 'Momentum'
+    params:
+        momentum: 0.9
+    regularizer:
+        function: 'L2'
+        factor: 0.000070
+
+TRAIN:
+    batch_size: 256
+    num_workers: 4
+    file_list: "./dataset/NUS-WIDE-SCENE/NUS-SCENE-dataset/multilabel_train_list.txt"
+    data_dir: "./dataset/NUS-WIDE-SCENE/NUS-SCENE-dataset/images"
+    shuffle_seed: 0
+    transforms:
+        - DecodeImage:
+            to_rgb: True
+            to_np: False
+            channel_first: False
+        - RandCropImage:
+            size: 224
+        - RandFlipImage:
+            flip_code: 1
+        - NormalizeImage:
+            scale: 1./255.
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: ''
+        - ToCHWImage:
+    mix:                       
+        - MixupOperator:    
+            alpha: 0.2      
+
+VALID:
+    batch_size: 64
+    num_workers: 4
+    file_list: "./dataset/NUS-WIDE-SCENE/NUS-SCENE-dataset/multilabel_test_list.txt"
+    data_dir: "./dataset/NUS-WIDE-SCENE/NUS-SCENE-dataset/images"
+    shuffle_seed: 0
+    transforms:
+        - DecodeImage:
+            to_rgb: True
+            to_np: False
+            channel_first: False
+        - ResizeImage:
+            resize_short: 256
+        - CropImage:
+            size: 224
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: ''
+        - ToCHWImage:
\ No newline at end of file

docs/en/advanced_tutorials/index.rst

@@ -6,4 +6,4 @@ advanced_tutorials
 
    image_augmentation/index
    distillation/index
-
+   multilabel/index

docs/en/advanced_tutorials/multilabel/index.rst

+Multilabel Classification
+================================
+
+.. toctree::
+   :maxdepth: 3
+
+   multilabel.md
\ No newline at end of file

docs/en/advanced_tutorials/multilabel/multilabel_en.md

+# Multilabel classification quick start
+
+Based on the [NUS-WIDE-SCENE](https://lms.comp.nus.edu.sg/wp-content/uploads/2019/research/nuswide/NUS-WIDE.html) dataset which is a subset of NUS-WIDE dataset, you can experience multilabel of PaddleClas, include training, evaluation and prediction. Please refer to [Installation](install.md) to install at first.
+
+## Preparation
+
+* Enter PaddleClas directory
+
+```
+cd path_to_PaddleClas
+```
+
+* Create and enter `dataset/NUS-WIDE-SCENE` directory, download and decompress NUS-WIDE-SCENE dataset
+
+```shell
+mkdir dataset/NUS-WIDE-SCENE
+cd dataset/NUS-WIDE-SCENE
+wget https://paddle-imagenet-models-name.bj.bcebos.com/data/NUS-SCENE-dataset.tar
+tar -xf NUS-SCENE-dataset.tar
+```
+
+* Return `PaddleClas` root home
+
+```
+cd ../../
+```
+
+## Environment
+
+### Download pretrained model
+
+You can use the following commands to download the pretrained model of ResNet50_vd.
+
+```bash
+mkdir pretrained
+cd pretrained
+wget https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNet50_vd_pretrained.pdparams
+cd ../
+```
+
+## Training
+
+```shell
+export CUDA_VISIBLE_DEVICES=0
+python -m paddle.distributed.launch \
+    --gpus="0" \
+    tools/train.py \
+        -c ./configs/quick_start/ResNet50_vd_multilabel.yaml
+```
+
+After training for 10 epochs, the best accuracy over the validation set should be around 0.72.
+
+## Evaluation
+
+```bash
+python tools/eval.py \
+    -c ./configs/quick_start/ResNet50_vd_multilabel.yaml \
+    -o pretrained_model="./output/ResNet50_vd/best_model/ppcls" \
+    -o load_static_weights=False
+```
+
+The metric of evaluation is based on mAP, which is commonly used in multilabel task to show model perfermance. The mAP over validation set should be around 0.57.
+
+## Prediction
+
+```bash
+python tools/infer/infer.py \
+    -i "./dataset/NUS-WIDE-SCENE/NUS-SCENE-dataset/images/0199_434752251.jpg" \
+    --model ResNet50_vd \
+    --pretrained_model "./output/ResNet50_vd/best_model/ppcls" \
+    --use_gpu True \
+    --load_static_weights False \
+    --multilabel True \
+    --class_num 33
+```
+
+You will get multiple output such as the following:
+```    
+    class id: 3, probability: 0.6025
+    class id: 23, probability: 0.5491
+    class id: 32, probability: 0.7006
+```
\ No newline at end of file

docs/zh_CN/advanced_tutorials/index.rst

@@ -6,4 +6,4 @@
 
    image_augmentation/index
    distillation/index
-
+   multilabel/index

docs/zh_CN/advanced_tutorials/multilabel/index.rst

+多标签分类
+================================
+
+.. toctree::
+   :maxdepth: 3
+
+   multilabel.md
\ No newline at end of file

docs/zh_CN/advanced_tutorials/multilabel/multilabel.md

+# 多标签分类quick start
+
+基于[NUS-WIDE-SCENE](https://lms.comp.nus.edu.sg/wp-content/uploads/2019/research/nuswide/NUS-WIDE.html)数据集，体验多标签分类的训练、评估、预测的过程，该数据集是NUS-WIDE数据集的一个子集。请事先参考[安装指南](install.md)配置运行环境和克隆PaddleClas代码。
+
+## 一、数据和模型准备
+
+* 进入PaddleClas目录。
+
+```
+cd path_to_PaddleClas
+```
+
+* 创建并进入`dataset/NUS-WIDE-SCENE`目录，下载并解压NUS-WIDE-SCENE数据集。
+
+```shell
+mkdir dataset/NUS-WIDE-SCENE
+cd dataset/NUS-WIDE-SCENE
+wget https://paddle-imagenet-models-name.bj.bcebos.com/data/NUS-SCENE-dataset.tar
+tar -xf NUS-SCENE-dataset.tar
+```
+
+* 返回`PaddleClas`根目录
+
+```
+cd ../../
+```
+
+## 二、环境准备
+
+### 2.1 下载预训练模型
+
+本例展示基于ResNet50_vd模型的多标签分类流程，因此首先下载ResNet50_vd的预训练模型
+
+```bash
+mkdir pretrained
+cd pretrained
+wget https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNet50_vd_pretrained.pdparams
+cd ../
+```
+
+## 三、模型训练
+
+```shell
+export CUDA_VISIBLE_DEVICES=0
+python -m paddle.distributed.launch \
+    --gpus="0" \
+    tools/train.py \
+        -c ./configs/quick_start/ResNet50_vd_multilabel.yaml
+```
+
+训练10epoch之后，验证集最好的正确率应该在0.72左右。
+
+## 四、模型评估
+
+```bash
+python tools/eval.py \
+    -c ./configs/quick_start/ResNet50_vd_multilabel.yaml \
+    -o pretrained_model="./output/ResNet50_vd/best_model/ppcls" \
+    -o load_static_weights=False
+```
+
+评估指标采用mAP，验证集的mAP应该在0.57左右。
+
+## 五、模型预测
+
+```bash
+python tools/infer/infer.py \
+    -i "./dataset/NUS-WIDE-SCENE/NUS-SCENE-dataset/images/0199_434752251.jpg" \
+    --model ResNet50_vd \
+    --pretrained_model "./output/ResNet50_vd/best_model/ppcls" \
+    --use_gpu True \
+    --load_static_weights False \
+    --multilabel True \
+    --class_num 33
+```
+
+得到类似下面的输出：
+```    
+    class id: 3, probability: 0.6025
+    class id: 23, probability: 0.5491
+    class id: 32, probability: 0.7006
+```
\ No newline at end of file

ppcls/data/reader.py

@@ -197,6 +197,40 @@ class CommonDataset(Dataset):
 
     def __len__(self):
         return self.num_samples
+    
+
+class MultiLabelDataset(Dataset):
+    """
+    Define dataset class for multilabel image classification
+    """
+
+    def __init__(self, params):
+        self.params = params
+        self.mode = params.get("mode", "train")
+        self.full_lines = get_file_list(params)
+        self.delimiter = params.get("delimiter", "\t")
+        self.ops = create_operators(params["transforms"])
+        self.num_samples = len(self.full_lines)
+        return
+
+    def __getitem__(self, idx):
+        try:
+            line = self.full_lines[idx]
+            img_path, label_str = line.split(self.delimiter)
+            img_path = os.path.join(self.params["data_dir"], img_path)
+            with open(img_path, "rb") as f:
+                img = f.read()
+
+            labels = label_str.split(',')
+            labels = [int(i) for i in labels]
+
+            return (transform(img, self.ops), np.array(labels).astype("float32"))
+        except Exception as e:
+            logger.error("data read failed: {}, exception info: {}".format(line, e))
+            return self.__getitem__(random.randint(0, len(self)))
+
+    def __len__(self):
+        return self.num_samples
 
 
 class Reader:
@@ -229,6 +263,7 @@ class Reader:
             self.collate_fn = self.mix_collate_fn
 
         self.places = places
+        self.multilabel = config.get("multilabel", False)
 
     def mix_collate_fn(self, batch):
         batch = transform(batch, self.batch_ops)
@@ -246,7 +281,10 @@ class Reader:
     def __call__(self):
         batch_size = int(self.params['batch_size']) // trainers_num
 
-        dataset = CommonDataset(self.params)
+        if self.multilabel:
+            dataset = MultiLabelDataset(self.params)
+        else:
+            dataset = CommonDataset(self.params)
 
         is_train = self.params['mode'] == "train"
         batch_sampler = DistributedBatchSampler(

ppcls/modeling/loss.py

@@ -15,7 +15,7 @@
 import paddle
 import paddle.nn.functional as F
 
-__all__ = ['CELoss', 'MixCELoss', 'GoogLeNetLoss', 'JSDivLoss']
+__all__ = ['CELoss', 'MixCELoss', 'GoogLeNetLoss', 'JSDivLoss', 'MultiLabelLoss']
 
 
 class Loss(object):
@@ -41,6 +41,17 @@ class Loss(object):
         soft_target = F.label_smooth(one_hot_target, epsilon=self._epsilon)
         soft_target = paddle.reshape(soft_target, shape=[-1, self._class_dim])
         return soft_target
+    
+    def _binary_crossentropy(self, input, target):
+        if self._label_smoothing:
+            target = self._labelsmoothing(target)
+            cost = F.binary_cross_entropy_with_logits(logit=input, label=target)
+        else:
+            cost = F.binary_cross_entropy_with_logits(logit=input, label=target)
+
+        avg_cost = paddle.mean(cost)
+
+        return avg_cost
 
     def _crossentropy(self, input, target):
         if self._label_smoothing:
@@ -68,6 +79,20 @@ class Loss(object):
 
     def __call__(self, input, target):
         pass
+    
+    
+class MultiLabelLoss(Loss):
+    """
+    Multilabel loss based binary cross entropy
+    """
+
+    def __init__(self, class_dim=1000, epsilon=None):
+        super(MultiLabelLoss, self).__init__(class_dim, epsilon)
+
+    def __call__(self, input, target):
+        cost = self._binary_crossentropy(input, target)
+
+        return cost
 
 
 class CELoss(Loss):

ppcls/utils/__init__.py

@@ -15,7 +15,13 @@
 from . import logger
 from . import misc
 from . import model_zoo
+from . import metrics
 
 from .save_load import init_model, save_model
 from .config import get_config
 from .misc import AverageMeter
+from .metrics import multi_hot_encode
+from .metrics import hamming_distance
+from .metrics import accuracy_score
+from .metrics import precision_recall_fscore
+from .metrics import mean_average_precision

ppcls/utils/metrics.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# 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.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+from sklearn.metrics import hamming_loss
+from sklearn.metrics import accuracy_score as accuracy_metric
+from sklearn.metrics import multilabel_confusion_matrix
+from sklearn.metrics import precision_recall_fscore_support
+from sklearn.metrics import average_precision_score
+from sklearn.preprocessing import binarize
+
+import numpy as np
+
+__all__ = ["multi_hot_encode", "hamming_distance", "accuracy_score", "precision_recall_fscore", "mean_average_precision"]
+
+
+def multi_hot_encode(logits, threshold=0.5):
+    """
+    Encode logits to multi-hot by elementwise for multilabel
+    """
+
+    return binarize(logits, threshold)
+
+
+def hamming_distance(output, target):
+    """
+    Soft metric based label for multilabel classification
+    Returns:
+        The smaller the return value is, the better model is.
+    """
+
+    return hamming_loss(target, output)
+
+
+def accuracy_score(output, target, base="sample"):
+    """
+    Hard metric for multilabel classification
+    Args:
+        output:
+        target:
+        base: ["sample", "label"], default="sample"
+            if "sample", return metric score based sample,
+            if "label", return metric score based label.
+    Returns:
+        accuracy:
+    """
+
+    assert base in ["sample", "label"], 'must be one of ["sample", "label"]'
+
+    if base == "sample":
+        accuracy = accuracy_metric(target, output)
+    elif base == "label":
+        mcm = multilabel_confusion_matrix(target, output)
+        tns = mcm[:, 0, 0]
+        fns = mcm[:, 1, 0]
+        tps = mcm[:, 1, 1]
+        fps = mcm[:, 0, 1]
+
+        accuracy = (sum(tps) + sum(tns)) / (sum(tps) + sum(tns) + sum(fns) + sum(fps))
+
+    return accuracy
+
+
+def precision_recall_fscore(output, target):
+    """
+    Metric based label for multilabel classification
+    Returns:
+        precisions:
+        recalls:
+        fscores:
+    """
+
+    precisions, recalls, fscores, _ = precision_recall_fscore_support(target, output)
+
+    return precisions, recalls, fscores
+
+
+def mean_average_precision(logits, target):
+    """
+    Calculate average precision
+    Args:
+        logits: probability from network before sigmoid or softmax
+        target: ground truth, 0 or 1
+    """
+    if not (isinstance(logits, np.ndarray) and isinstance(target, np.ndarray)):
+        raise TypeError("logits and target should be np.ndarray.")
+
+    aps = []
+    for i in range(target.shape[1]):
+        ap = average_precision_score(target[:, i], logits[:, i])
+        aps.append(ap)
+
+    return np.mean(aps)

requirements.txt

@@ -5,3 +5,4 @@ tqdm
 PyYAML
 visualdl >= 2.0.0b
 scipy
+scikit-learn==0.23.2

tools/eval.py

@@ -13,6 +13,7 @@
 # limitations under the License.
 
 import paddle
+import paddle.nn.functional as F
 
 import argparse
 import os
@@ -24,9 +25,15 @@ sys.path.append(os.path.abspath(os.path.join(__dir__, '..')))
 from ppcls.utils import logger
 from ppcls.utils.save_load import init_model
 from ppcls.utils.config import get_config
+from ppcls.utils import multi_hot_encode
+from ppcls.utils import accuracy_score
+from ppcls.utils import mean_average_precision
+from ppcls.utils import precision_recall_fscore
 from ppcls.data import Reader
 import program
 
+import numpy as np
+
 
 def parse_args():
     parser = argparse.ArgumentParser("PaddleClas eval script")
@@ -52,6 +59,7 @@ def main(args, return_dict={}):
     # assign place
     use_gpu = config.get("use_gpu", True)
     place = paddle.set_device('gpu' if use_gpu else 'cpu')
+    multilabel = config.get("multilabel", False)
 
     trainer_num = paddle.distributed.get_world_size()
     use_data_parallel = trainer_num != 1
@@ -68,12 +76,38 @@ def main(args, return_dict={}):
     valid_dataloader = Reader(config, 'valid', places=place)()
     net.eval()
     with paddle.no_grad():
-        top1_acc = program.run(valid_dataloader, config, net, None, None, 0,
-                               'valid')
-    return_dict["top1_acc"] = top1_acc
-    return top1_acc
+        if not multilabel:
+            top1_acc = program.run(valid_dataloader, config, net, None, None, 0,
+                                   'valid')
+            return_dict["top1_acc"] = top1_acc
+
+            return top1_acc
+        else:
+            all_outs = []
+            targets = []
+            for idx, batch in enumerate(valid_dataloader()):
+                feeds = program.create_feeds(batch, False, config.classes_num, multilabel)
+                out = net(feeds["image"])
+                out = F.sigmoid(out)
+
+                use_distillation = config.get("use_distillation", False)
+                if use_distillation:
+                    out = out[1]
+
+                all_outs.extend(list(out.numpy()))
+                targets.extend(list(feeds["label"].numpy()))
+            all_outs = np.array(all_outs)
+            targets = np.array(targets)
+
+            mAP = mean_average_precision(all_outs, targets)
+
+            return_dict["mean average precision"] = mAP
+
+            return mAP
 
 
 if __name__ == '__main__':
     args = parse_args()
-    main(args)
+    return_dict = {}
+    main(args, return_dict)
+    print(return_dict)

tools/infer/infer.py

@@ -34,6 +34,7 @@ def main():
     args = parse_args()
     # assign the place
     place = paddle.set_device('gpu' if args.use_gpu else 'cpu')
+    multilabel = True if args.multilabel else False
 
     net = architectures.__dict__[args.model](class_dim=args.class_num)
     load_dygraph_pretrain(net, args.pretrained_model, args.load_static_weights)
@@ -61,17 +62,25 @@ def main():
             batch_outputs = net(batch_tensor)
             if args.model == "GoogLeNet":
                 batch_outputs = batch_outputs[0]
-            batch_outputs = F.softmax(batch_outputs)
+            if multilabel:
+                batch_outputs = F.sigmoid(batch_outputs)
+            else:
+                batch_outputs = F.softmax(batch_outputs)
             batch_outputs = batch_outputs.numpy()
-            batch_result_list = postprocess(batch_outputs, args.top_k)
+            batch_result_list = postprocess(batch_outputs, args.top_k, multilabel=multilabel)
 
             for number, result_dict in enumerate(batch_result_list):
                 filename = img_path_list[number].split("/")[-1]
                 clas_ids = result_dict["clas_ids"]
-                scores_str = "[{}]".format(", ".join("{:.2f}".format(
-                    r) for r in result_dict["scores"]))
-                print("File:{}, Top-{} result: class id(s): {}, score(s): {}".
-                      format(filename, args.top_k, clas_ids, scores_str))
+                if multilabel:
+                    print("File:{}, multilabel result: ".format(filename))
+                    for id, score in zip(clas_ids, result_dict["scores"]):
+                        print("\tclass id: {}, probability: {:.2f}".format(id, score))
+                else:
+                    scores_str = "[{}]".format(", ".join("{:.2f}".format(
+                        r) for r in result_dict["scores"]))
+                    print("File:{}, Top-{} result: class id(s): {}, score(s): {}".
+                        format(filename, args.top_k, clas_ids, scores_str))
 
                 if args.pre_label_image:
                     save_prelabel_results(clas_ids[0], img_path_list[number],

tools/infer/utils.py

@@ -31,6 +31,7 @@ def parse_args():
     parser = argparse.ArgumentParser()
     parser.add_argument("-i", "--image_file", type=str)
     parser.add_argument("--use_gpu", type=str2bool, default=True)
+    parser.add_argument("--multilabel", type=str2bool, default=False)
 
     # params for preprocess
     parser.add_argument("--resize_short", type=int, default=256)
@@ -124,11 +125,14 @@ def preprocess(img, args):
     return img
 
 
-def postprocess(batch_outputs, topk=5):
+def postprocess(batch_outputs, topk=5, multilabel=False):
     batch_results = []
     for probs in batch_outputs:
         results = []
-        index = probs.argsort(axis=0)[-topk:][::-1].astype("int32")
+        if multilabel:
+            index = np.where(probs >= 0.5)[0].astype('int32')
+        else:
+            index = probs.argsort(axis=0)[-topk:][::-1].astype("int32")
         clas_id_list = []
         score_list = []
         for i in index:

tools/program.py

@@ -29,12 +29,16 @@ import paddle.nn.functional as F
 from ppcls.optimizer import LearningRateBuilder
 from ppcls.optimizer import OptimizerBuilder
 from ppcls.modeling import architectures
+from ppcls.modeling.loss import MultiLabelLoss
 from ppcls.modeling.loss import CELoss
 from ppcls.modeling.loss import MixCELoss
 from ppcls.modeling.loss import JSDivLoss
 from ppcls.modeling.loss import GoogLeNetLoss
 from ppcls.utils.misc import AverageMeter
 from ppcls.utils import logger
+from ppcls.utils import multi_hot_encode
+from ppcls.utils import hamming_distance
+from ppcls.utils import accuracy_score
 
 
 def create_model(architecture, classes_num):
@@ -61,7 +65,8 @@ def create_loss(feeds,
                 classes_num=1000,
                 epsilon=None,
                 use_mix=False,
-                use_distillation=False):
+                use_distillation=False,
+                multilabel=False):
     """
     Create a loss for optimization, such as:
         1. CrossEnotry loss
@@ -100,7 +105,10 @@ def create_loss(feeds,
         feed_lam = feeds['lam']
         return loss(out, feed_y_a, feed_y_b, feed_lam)
     else:
-        loss = CELoss(class_dim=classes_num, epsilon=epsilon)
+        if not multilabel:
+            loss = CELoss(class_dim=classes_num, epsilon=epsilon)
+        else:
+            loss = MultiLabelLoss(class_dim=classes_num, epsilon=epsilon)
         return loss(out, feeds["label"])
 
 
@@ -110,6 +118,7 @@ def create_metric(out,
                   topk=5,
                   classes_num=1000,
                   use_distillation=False,
+                  multilabel=False,
                   mode="train"):
     """
     Create measures of model accuracy, such as top1 and top5
@@ -135,24 +144,43 @@ def create_metric(out,
     softmax_out = F.softmax(out)
 
     fetchs = OrderedDict()
-    # set top1 to fetchs
-    top1 = paddle.metric.accuracy(softmax_out, label=label, k=1)
-    # set topk to fetchs
-    k = min(topk, classes_num)
-    topk = paddle.metric.accuracy(softmax_out, label=label, k=k)
+    metric_names = set()
+    if not multilabel:
+        softmax_out = F.softmax(out)
+
+        # set top1 to fetchs
+        top1 = paddle.metric.accuracy(softmax_out, label=label, k=1)
+        # set topk to fetchs
+        k = min(topk, classes_num)
+        topk = paddle.metric.accuracy(softmax_out, label=label, k=k)
+
+        metric_names.add("top1")
+        metric_names.add("top{}".format(k))
+
+        fetchs['top1'] = top1
+        topk_name = "top{}".format(k)
+        fetchs[topk_name] = topk
+    else:
+        out = F.sigmoid(out)
+        preds = multi_hot_encode(out.numpy())
+        targets = label.numpy()
+        ham_dist = to_tensor(hamming_distance(preds, targets))
+        accuracy = to_tensor(accuracy_score(preds, targets, base="label"))
+
+        ham_dist_name = "hamming_distance"
+        accuracy_name = "multilabel_accuracy"
+        metric_names.add(ham_dist_name)
+        metric_names.add(accuracy_name)
+
+        fetchs[accuracy_name] = accuracy
+        fetchs[ham_dist_name] = ham_dist
 
     # multi cards' eval
     if mode != "train" and paddle.distributed.get_world_size() > 1:
-        top1 = paddle.distributed.all_reduce(
-            top1, op=paddle.distributed.ReduceOp.
-            SUM) / paddle.distributed.get_world_size()
-        topk = paddle.distributed.all_reduce(
-            topk, op=paddle.distributed.ReduceOp.
-            SUM) / paddle.distributed.get_world_size()
-
-    fetchs['top1'] = top1
-    topk_name = 'top{}'.format(k)
-    fetchs[topk_name] = topk
+        for metric_name in metric_names:
+            fetchs[metric_name] = paddle.distributed.all_reduce(
+                fetchs[metric_name], op=paddle.distributed.ReduceOp.
+                SUM) / paddle.distributed.get_world_size()
 
     return fetchs
 
@@ -182,12 +210,14 @@ def create_fetchs(feeds, net, config, mode="train"):
     epsilon = config.get('ls_epsilon')
     use_mix = config.get('use_mix') and mode == 'train'
     use_distillation = config.get('use_distillation')
+    multilabel = config.get('multilabel', False)
 
     out = net(feeds["image"])
 
     fetchs = OrderedDict()
     fetchs['loss'] = create_loss(feeds, out, architecture, classes_num,
-                                 epsilon, use_mix, use_distillation)
+                                 epsilon, use_mix, use_distillation,
+                                 multilabel)
     if not use_mix:
         metric = create_metric(
             out,
@@ -196,6 +226,7 @@ def create_fetchs(feeds, net, config, mode="train"):
             topk,
             classes_num,
             use_distillation,
+            multilabel=multilabel,
             mode=mode)
         fetchs.update(metric)
 
@@ -240,7 +271,7 @@ def create_optimizer(config, parameter_list=None):
     return opt(lr, parameter_list), lr
 
 
-def create_feeds(batch, use_mix):
+def create_feeds(batch, use_mix, num_classes, multilabel=False):
     image = batch[0]
     if use_mix:
         y_a = to_tensor(batch[1].numpy().astype("int64").reshape(-1, 1))
@@ -248,7 +279,10 @@ def create_feeds(batch, use_mix):
         lam = to_tensor(batch[3].numpy().astype("float32").reshape(-1, 1))
         feeds = {"image": image, "y_a": y_a, "y_b": y_b, "lam": lam}
     else:
-        label = to_tensor(batch[1].numpy().astype('int64').reshape(-1, 1))
+        if not multilabel:
+            label = to_tensor(batch[1].numpy().astype("int64").reshape(-1, 1))
+        else:
+            label = to_tensor(batch[1].numpy().astype('float32').reshape(-1, num_classes))
         feeds = {"image": image, "label": label}
     return feeds
 
@@ -279,6 +313,8 @@ def run(dataloader,
     """
     print_interval = config.get("print_interval", 10)
     use_mix = config.get("use_mix", False) and mode == "train"
+    multilabel = config.get("multilabel", False)
+    classes_num = config.get("classes_num")
 
     metric_list = [
         ("loss", AverageMeter(
@@ -291,13 +327,19 @@ def run(dataloader,
             'reader_cost', '.5f', postfix=" s,")),
     ]
     if not use_mix:
-        topk_name = 'top{}'.format(config.topk)
-        metric_list.insert(
-            0, (topk_name, AverageMeter(
-                topk_name, '.5f', postfix=",")))
-        metric_list.insert(
-            0, ("top1", AverageMeter(
-                "top1", '.5f', postfix=",")))
+        if not multilabel:
+            topk_name = 'top{}'.format(config.topk)
+            metric_list.insert(
+                0, (topk_name, AverageMeter(
+                    topk_name, '.5f', postfix=",")))
+            metric_list.insert(
+                0, ("top1", AverageMeter(
+                    "top1", '.5f', postfix=",")))
+        else:
+            metric_list.insert(0, ("multilabel_accuracy", AverageMeter(
+                                   "multilabel_accuracy", '.5f', postfix=",")))
+            metric_list.insert(0, ("hamming_distance", AverageMeter(
+                                   "hamming_distance", '.5f', postfix=",")))
 
     metric_list = OrderedDict(metric_list)
 
@@ -310,7 +352,7 @@ def run(dataloader,
 
         metric_list['reader_time'].update(time.time() - tic)
         batch_size = len(batch[0])
-        feeds = create_feeds(batch, use_mix)
+        feeds = create_feeds(batch, use_mix, classes_num, multilabel)
         fetchs = create_fetchs(feeds, net, config, mode)
         if mode == 'train':
             avg_loss = fetchs['loss']
@@ -387,4 +429,7 @@ def run(dataloader,
 
     # return top1_acc in order to save the best model
     if mode == 'valid':
-        return metric_list['top1'].avg
+        if multilabel:
+            return metric_list['multilabel_accuracy'].avg
+        else:
+            return metric_list['top1'].avg