add weighted soft labels loss function

user3984 · user3984 · commit 0cfccd769d19 · 2022-09-14T04:48:42.000Z
diff --git a/docs/zh_CN/advanced_tutorials/knowledge_distillation.md b/docs/zh_CN/advanced_tutorials/knowledge_distillation.md
@@ -16,6 +16,7 @@
         - [1.2.5 DKD](#1.2.5)
         - [1.2.6 DIST](#1.2.6)
         - [1.2.7 MGD](#1.2.7)
+        - [1.2.8 WSL](#1.2.8)
 - [2. 使用方法](#2)
     - [2.1 环境配置](#2.1)
     - [2.2 数据准备](#2.2)
@@ -399,7 +400,7 @@ DKD将蒸馏中常用的 KD Loss 进行了解耦成为Target Class Knowledge Dis
 | 策略 | 骨干网络 | 配置文件 | Top-1 acc | 下载链接 |
 | --- | --- | --- | --- | --- |
 | baseline | ResNet18 | [ResNet18.yaml](../../../ppcls/configs/ImageNet/ResNet/ResNet18.yaml) | 70.8% | - |
-| AFD | ResNet18 | [resnet34_distill_resnet18_dkd.yaml](../../../ppcls/configs/ImageNet/Distillation/resnet34_distill_resnet18_dkd.yaml) | 72.59%(**+1.79%**) | - |
+| DKD | ResNet18 | [resnet34_distill_resnet18_dkd.yaml](../../../ppcls/configs/ImageNet/Distillation/resnet34_distill_resnet18_dkd.yaml) | 72.59%(**+1.79%**) | - |
 
 
 ##### 1.2.5.2 DKD 配置
@@ -583,6 +584,73 @@ Loss:
         weight: 1.0
 ```
 
+<a name='1.2.8'></a>
+
+#### 1.2.8 WSL
+
+##### 1.2.8.1 WSL 算法介绍
+
+论文信息：
+
+
+> [Rethinking Soft Labels For Knowledge Distillation: A Bias-variance Tradeoff Perspective](https://arxiv.org/abs/2102.0650)
+>
+> Helong Zhou, Liangchen Song, Jiajie Chen, Ye Zhou, Guoli Wang, Junsong Yuan, Qian Zhang
+>
+> ICLR, 2021
+
+WSL (Weighted Soft Labels) 损失函数根据教师模型与学生模型关于真值标签的 CE Loss 比值，对每个样本的 KD Loss 分别赋予权重。若学生模型相对教师模型在某个样本上预测结果更好，则对该样本赋予较小的权重。该方法简单、有效，使各个样本的权重可自适应调节，提升了蒸馏精度。
+
+在ImageNet1k公开数据集上，效果如下所示。
+
+| 策略 | 骨干网络 | 配置文件 | Top-1 acc | 下载链接 |
+| --- | --- | --- | --- | --- |
+| baseline | ResNet18 | [ResNet18.yaml](../../../ppcls/configs/ImageNet/ResNet/ResNet18.yaml) | 70.8% | - |
+| WSL | ResNet18 | [resnet34_distill_resnet18_wsl.yaml](../../../ppcls/configs/ImageNet/Distillation/esnet34_distill_resnet18_wsl.yaml) | 72.23%(**+1.43%**) | - |
+
+
+##### 1.2.8.2 WSL 配置
+
+WSL 配置如下所示。在模型构建Arch字段中，需要同时定义学生模型与教师模型，教师模型固定参数，且需要加载预训练模型。在损失函数Loss字段中，需要定义`DistillationGTCELoss`（学生与真值标签之间的CE loss）以及`DistillationWSLLoss`（学生与教师之间的WSL loss），作为训练的损失函数。
+
+
+```yaml
+# model architecture
+Arch:
+  name: "DistillationModel"
+  # if not null, its lengths should be same as models
+  pretrained_list:
+  # if not null, its lengths should be same as models
+  freeze_params_list:
+  - True
+  - False
+  models:
+    - Teacher:
+        name: ResNet34
+        pretrained: True
+
+    - Student:
+        name: ResNet18
+        pretrained: False
+
+  infer_model_name: "Student"
+
+
+# loss function config for traing/eval process
+Loss:
+  Train:
+    - DistillationGTCELoss:
+        weight: 1.0
+        model_names: ["Student"]
+    - DistillationWSLLoss:
+        weight: 2.5
+        model_name_pairs: [["Student", "Teacher"]]
+        temperature: 2
+  Eval:
+    - CELoss:
+        weight: 1.0
+```
+
 <a name="2"></a>
 
 ## 2. 模型训练、评估和预测
diff --git a/ppcls/configs/ImageNet/Distillation/resnet34_distill_resnet18_wsl.yaml b/ppcls/configs/ImageNet/Distillation/resnet34_distill_resnet18_wsl.yaml
@@ -0,0 +1,152 @@
+# global configs
+Global:
+  checkpoints: null
+  pretrained_model: null
+  output_dir: ./output/r34_r18_wsl
+  device: "gpu"
+  save_interval: 1
+  eval_during_train: True
+  eval_interval: 1
+  epochs: 100
+  print_batch_step: 10
+  use_visualdl: False
+  # used for static mode and model export
+  image_shape: [3, 224, 224]
+  save_inference_dir: "./inference"
+
+# model architecture
+Arch:
+  name: "DistillationModel"
+  # if not null, its lengths should be same as models
+  pretrained_list:
+  # if not null, its lengths should be same as models
+  freeze_params_list:
+  - True
+  - False
+  models:
+    - Teacher:
+        name: ResNet34
+        pretrained: True
+
+    - Student:
+        name: ResNet18
+        pretrained: False
+
+  infer_model_name: "Student"
+
+
+# loss function config for traing/eval process
+Loss:
+  Train:
+    - DistillationGTCELoss:
+        weight: 1.0
+        model_names: ["Student"]
+    - DistillationWSLLoss:
+        weight: 2.5
+        model_name_pairs: [["Student", "Teacher"]]
+        temperature: 2
+  Eval:
+    - CELoss:
+        weight: 1.0
+
+
+Optimizer:
+  name: Momentum
+  momentum: 0.9
+  weight_decay: 1e-4
+  lr:
+    name: MultiStepDecay
+    learning_rate: 0.1
+    milestones: [30, 60, 90]
+    step_each_epoch: 1
+    gamma: 0.1
+
+
+# data loader for train and eval
+DataLoader:
+  Train:
+    dataset:
+        name: ImageNetDataset
+        image_root: "./dataset/ILSVRC2012/"
+        cls_label_path: "./dataset/ILSVRC2012/train_list.txt"
+        transform_ops:
+          - DecodeImage:
+              to_rgb: True
+              channel_first: False
+          - RandCropImage:
+              size: 224
+          - RandFlipImage:
+              flip_code: 1
+          - NormalizeImage:
+              scale: 0.00392157
+              mean: [0.485, 0.456, 0.406]
+              std: [0.229, 0.224, 0.225]
+              order: ''
+
+    sampler:
+        name: DistributedBatchSampler
+        batch_size: 64
+        drop_last: False
+        shuffle: True
+    loader:
+        num_workers: 8
+        use_shared_memory: True
+
+  Eval:
+    dataset: 
+        name: ImageNetDataset
+        image_root: "./dataset/ILSVRC2012/"
+        cls_label_path: "./dataset/ILSVRC2012/val_list.txt"
+        transform_ops:
+          - DecodeImage:
+              to_rgb: True
+              channel_first: False
+          - ResizeImage:
+              resize_short: 256
+          - CropImage:
+              size: 224
+          - NormalizeImage:
+              scale: 0.00392157
+              mean: [0.485, 0.456, 0.406]
+              std: [0.229, 0.224, 0.225]
+              order: ''
+    sampler:
+        name: DistributedBatchSampler
+        batch_size: 64
+        drop_last: False
+        shuffle: False
+    loader:
+        num_workers: 4
+        use_shared_memory: True
+
+Infer:
+  infer_imgs: "docs/images/inference_deployment/whl_demo.jpg"
+  batch_size: 10
+  transforms:
+      - DecodeImage:
+          to_rgb: True
+          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:
+  PostProcess:
+    name: Topk
+    topk: 5
+    class_id_map_file: "ppcls/utils/imagenet1k_label_list.txt"
+
+Metric:
+    Train:
+    - DistillationTopkAcc:
+        model_key: "Student"
+        topk: [1, 5]
+    Eval:
+    - DistillationTopkAcc:
+        model_key: "Student"
+        topk: [1, 5]
diff --git a/ppcls/loss/__init__.py b/ppcls/loss/__init__.py
@@ -25,6 +25,7 @@
 from .distillationloss import DistillationRKDLoss
 from .distillationloss import DistillationKLDivLoss
 from .distillationloss import DistillationDKDLoss
+from .distillationloss import DistillationWSLLoss
 from .distillationloss import DistillationMultiLabelLoss
 from .distillationloss import DistillationDISTLoss
 from .distillationloss import DistillationPairLoss
diff --git a/ppcls/loss/distillationloss.py b/ppcls/loss/distillationloss.py
@@ -22,6 +22,7 @@
 from .rkdloss import RKdAngle, RkdDistance
 from .kldivloss import KLDivLoss
 from .dkdloss import DKDLoss
+from .wslloss import WSLLoss
 from .dist_loss import DISTLoss
 from .multilabelloss import MultiLabelLoss
 from .mgd_loss import MGDLoss
@@ -262,6 +263,34 @@ def forward(self, predicts, batch):
         return loss_dict
 
 
+class DistillationWSLLoss(WSLLoss):
+    """
+    DistillationWSLLoss
+    """
+
+    def __init__(self,
+                 model_name_pairs=[],
+                 key=None,
+                 temperature=2.0,
+                 name="wsl_loss"):
+        super().__init__(temperature)
+        self.model_name_pairs = model_name_pairs
+        self.key = key
+        self.name = name
+
+    def forward(self, predicts, batch):
+        loss_dict = dict()
+        for idx, pair in enumerate(self.model_name_pairs):
+            out1 = predicts[pair[0]]
+            out2 = predicts[pair[1]]
+            if self.key is not None:
+                out1 = out1[self.key]
+                out2 = out2[self.key]
+            loss = super().forward(out1, out2, batch)
+            loss_dict[f"{self.name}_{pair[0]}_{pair[1]}"] = loss
+        return loss_dict
+
+
 class DistillationMultiLabelLoss(MultiLabelLoss):
     """
     DistillationMultiLabelLoss
diff --git a/ppcls/loss/wslloss.py b/ppcls/loss/wslloss.py
@@ -0,0 +1,66 @@
+# copyright (c) 2022 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.
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+
+class WSLLoss(nn.Layer):
+    """
+    Weighted Soft Labels Loss
+    paper: https://arxiv.org/pdf/2102.00650.pdf
+    code reference: https://github.com/bellymonster/Weighted-Soft-Label-Distillation
+    """
+
+    def __init__(self, temperature=2.0, use_target_as_gt=False):
+        super().__init__()
+        self.temperature = temperature
+        self.use_target_as_gt = use_target_as_gt
+
+    def forward(self, logits_student, logits_teacher, target=None):
+        """Compute weighted soft labels loss.
+        Args:
+            logits_student: student's logits with shape (batch_size, num_classes)
+            logits_teacher: teacher's logits with shape (batch_size, num_classes)
+            target: ground truth labels with shape (batch_size)
+        """
+        if target is None or self.use_target_as_gt:
+            target = logits_teacher.argmax(axis=-1)
+
+        target = F.one_hot(
+            target.reshape([-1]), num_classes=logits_student[0].shape[0])
+
+        s_input_for_softmax = logits_student / self.temperature
+        t_input_for_softmax = logits_teacher / self.temperature
+
+        ce_loss_s = -paddle.sum(target *
+                                F.log_softmax(logits_student.detach()),
+                                axis=1)
+        ce_loss_t = -paddle.sum(target *
+                                F.log_softmax(logits_teacher.detach()),
+                                axis=1)
+
+        ratio = ce_loss_s / (ce_loss_t + 1e-7)
+        ratio = paddle.maximum(ratio, paddle.zeros_like(ratio))
+
+        kd_loss = -paddle.sum(F.softmax(t_input_for_softmax) *
+                              F.log_softmax(s_input_for_softmax),
+                              axis=1)
+        weight = 1 - paddle.exp(-ratio)
+
+        weighted_kd_loss = (self.temperature**2) * paddle.mean(kd_loss *
+                                                               weight)
+
+        return weighted_kd_loss
