diff --git a/docs/zh_CN_tmp/image_recognition_pipeline/feature_extraction.md b/docs/zh_CN_tmp/image_recognition_pipeline/feature_extraction.md index cade1a89da22c8019e88bc9b25a3cfbd86917f03..168afa64b5e095cd0782248dc694071570b13677 100644 --- a/docs/zh_CN_tmp/image_recognition_pipeline/feature_extraction.md +++ b/docs/zh_CN_tmp/image_recognition_pipeline/feature_extraction.md @@ -78,7 +78,7 @@ Loss: # 4.训练、评估、推理 下面以`ppcls/configs/Products/ResNet50_vd_SOP.yaml`为例,介绍模型的训练、评估、推理过程 ## 4.1 数据准备 -首先,下载SOP数据集, 数据链接 +首先,下载SOP数据集, 数据链接: https://cvgl.stanford.edu/projects/lifted_struct/ ## 4.2 训练 - 单机单卡训练 @@ -96,7 +96,7 @@ python -m paddle.distributed.launch ## 4.3 评估 - 单卡评估 ``` -python tools/eval.py -c ppcls/configs/ResNet50_vd_SOP.yaml -o Global.pretrained_model = "output/ReModel/best_model" +python tools/eval.py -c ppcls/configs/ResNet50_vd_SOP.yaml -o Global.pretrained_model="output/ReModel/best_model" ``` - 多卡评估 ``` @@ -109,9 +109,9 @@ python -m paddle.distributed.launch 推理过程包括两个步骤: 1) 导出推理模型; 2) 获取特征向量 ### 4.4.1 导出推理模型 ``` -python tools/export_model -c ppcls/configs/ResNet50_vd_SOP.yaml -o Global.pretrained_model = "output/ReModel/best_model" +python tools/export_model -c ppcls/configs/ResNet50_vd_SOP.yaml -o Global.pretrained_model="output/ReModel/best_model" ``` -生成的推理模型位于inference目录,名字为inference.pd* +生成的推理模型位于`inference`目录,名字为`inference.pd*` ### 4.4.2 获取特征向量 ```