Merge pull request #985 from RainFrost1/develop

update vehicle fine-graned classfication

docs/en/application/feature_learning_en.md

@@ -17,3 +17,14 @@ The overall structure of feature learning is shown in the figure above, which ma
 ## 2 Config Description
 
 The feature learning config file description can be found in [yaml description](../tutorials/config_en.md).
+
+## 3 Pretrained Model
+
+The following are the pretrained models trained on different dataset.
+
+- Vehicle Fine-Grained Classification：[CompCars](https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/models/pretrain/vehicle_cls_ResNet50_CompCars_v1.1_pretrained.pdparams)
+- Vehicle ReID：[VERI-Wild](https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/models/pretrain/vehicle_reid_ResNet50_VERIWild_v1.0_pretrained.pdparams)
+- Cartoon Character Recognition：[iCartoon](https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/models/pretrain/cartoon_rec_ResNet50_iCartoon_v1.0_pretrained.pdparams)
+- Logo Recognition：[Logo 3K](https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/models/pretrain/logo_rec_ResNet50_Logo3K_v1.0_pretrained.pdparams)
+- Product Recognition： [Inshop](https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/models/pretrain/product_ResNet50_vd_Inshop_pretrained_v1.0.pdparams)、[Aliproduct](https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/models/pretrain/product_ResNet50_vd_Aliproduct_v1.0_pretrained.pdparams) 
+

docs/en/application/vehicle_recognition_en.md

@@ -40,8 +40,7 @@ In order to reduce the complexity of calculating feature distance in inference,
 
 ### 1.4 Metric Learning Losses
 
-- In vehicle ReID，[SupConLoss](../../../ppcls/loss/supconloss.py) , [ArcLoss](../../../ppcls/arch/gears/arcmargin.py) are used. The weight ratio of two losses is 1:1.
-- In vehicle fine-grained classification, [TtripLet Loss](../../../ppcls/loss/triplet.py), [ArcLoss](../../../ppcls/arch/gears/arcmargin.py) are used. The weight ratio of two losses is 1:1.
+In vehicle ReID and vehicle fine-grained classification，[SupConLoss](../../../ppcls/loss/supconloss.py) , [ArcLoss](../../../ppcls/arch/gears/arcmargin.py) are used. The weight ratio of two losses is 1:1.
 
 ## 2 Experiment
 
@@ -74,13 +73,13 @@ The images in the dataset mainly come from the network and monitoring  data. The
 
  It is worth noting that this dataset needs to generate labels  according to its own needs. For example, in this demo, vehicles of the  same model produced in different years are regarded as the same  category. Therefore, the total number of categories is 431.
 
-|           **Methods**           | Top1 Acc  |
-| :-----------------------------: | :-------: |
-|        ResNet101-swp[6]         |   97.6%   |
-|      Fine-Tuning DARTS[7]       |   95.9%   |
-|       Resnet50 + COOC[8]        |   95.6%   |
-|             A3M[9]              |   95.4%   |
-| PaddleClas  baseline (ResNet50) | **97.1**% |
+|           **Methods**           |  Top1 Acc  |
+| :-----------------------------: | :--------: |
+|        ResNet101-swp[6]         |   97.6%    |
+|      Fine-Tuning DARTS[7]       |   95.9%    |
+|       Resnet50 + COOC[8]        |   95.6%    |
+|             A3M[9]              |   95.4%    |
+| PaddleClas  baseline (ResNet50) | **97.36**% |
 
 ## 3 References
 

docs/zh_CN/application/feature_learning.md

@@ -17,3 +17,13 @@
 ## 2 配置文件说明
 
 配置文件说明详见[yaml配置文件说明文档](../tutorials/config.md)。其中模型结构配置，详见文档中**识别模型结构配置**部分。
+
+## 3 预训练模型
+
+以下为各应用在不同数据集下的预训练模型
+
+- 车辆细分类：[CompCars](https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/models/pretrain/vehicle_cls_ResNet50_CompCars_v1.1_pretrained.pdparams)
+- 车辆ReID：[VERI-Wild](https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/models/pretrain/vehicle_reid_ResNet50_VERIWild_v1.0_pretrained.pdparams)
+- 动漫人物识别：[iCartoon](https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/models/pretrain/cartoon_rec_ResNet50_iCartoon_v1.0_pretrained.pdparams)
+- Logo识别：[Logo3K](https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/models/pretrain/logo_rec_ResNet50_Logo3K_v1.0_pretrained.pdparams)
+- 商品识别： [Inshop](https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/models/pretrain/product_ResNet50_vd_Inshop_pretrained_v1.0.pdparams)、[Aliproduct](https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/models/pretrain/product_ResNet50_vd_Aliproduct_v1.0_pretrained.pdparams) 

docs/zh_CN/application/vehicle_recognition.md

@@ -39,8 +39,7 @@ ReID，也就是 Re-identification，其定义是利用算法，在图像库中
 
 ### 1.4 Metric Learning相关Loss的设置
 
-- 车辆ReID中，使用了[SupConLoss](../../../ppcls/loss/supconloss.py) + [ArcLoss](../../../ppcls/arch/gears/arcmargin.py)，其中权重比例为1:1
-- 车辆细分类，使用[TtripLet Loss](../../../ppcls/loss/triplet.py) + [ArcLoss](../../../ppcls/arch/gears/arcmargin.py)，其中权重比例为1:1
+车辆ReID及细粒度分类中，使用了[SupConLoss](../../../ppcls/loss/supconloss.py) + [ArcLoss](../../../ppcls/arch/gears/arcmargin.py)，其中权重比例为1:1
 
 ## 2 实验结果
 
@@ -72,13 +71,13 @@ ReID，也就是 Re-identification，其定义是利用算法，在图像库中
 数据集中图像主要来自网络和监控数据，其中网络数据包含163个汽车制造商、1716个汽车型号的汽车。共**136,726**张全车图像，**27,618**张部分车图像。其中网络汽车数据包含bounding box、视角、5个属性（最大速度、排量、车门数、车座数、汽车类型）。监控数据包含**50,000**张前视角图像。
 值得注意的是，此数据集中需要根据自己的需要生成不同的label，如本demo中，将不同年份生产的相同型号的车辆视为同一类，因此，类别总数为：431类。
 
-|           **Methods**           | Top1 Acc  |
-| :-----------------------------: | :-------: |
-|        ResNet101-swp[6]         |   97.6%   |
-|      Fine-Tuning DARTS[7]       |   95.9%   |
-|       Resnet50 + COOC[8]        |   95.6%   |
-|             A3M[9]              |   95.4%   |
-| PaddleClas  baseline (ResNet50) | **97.1**% |
+|           **Methods**           |  Top1 Acc  |
+| :-----------------------------: | :--------: |
+|        ResNet101-swp[6]         |   97.6%    |
+|      Fine-Tuning DARTS[7]       |   95.9%    |
+|       Resnet50 + COOC[8]        |   95.6%    |
+|             A3M[9]              |   95.4%    |
+| PaddleClas  baseline (ResNet50) | **97.37**% |
 
 ## 3 参考文献
 

ppcls/configs/Vehicle/ResNet50.yaml

@@ -40,9 +40,9 @@ Loss:
   Train:
     - CELoss:
         weight: 1.0
-    - TripletLossV2:
+    - SupConLoss:
         weight: 1.0
-        margin: 0.5
+        views: 2
   Eval:
     - CELoss:
         weight: 1.0