未验证 提交 ff2a26c7 编写于 作者: B Bin Lu 提交者: GitHub

Update getting_started_retrieval.md

上级 f057eebd
......@@ -24,7 +24,7 @@ PaddleClas目前支持的训练/评估环境如下:
```
python tools/train.py \
-c configs/quick_start/ResNet50_flowers_retrieval_finetune.yaml \
-c configs/quick_start/ResNet50_vd_finetune_retrieval.yaml \
-o pretrained_model="" \
-o use_gpu=True
```
......@@ -41,7 +41,7 @@ python tools/train.py \
```
python tools/train.py \
-c configs/quick_start/ResNet50_flowers_retrieval_finetune.yaml \
-c configs/quick_start/ResNet50_vd_finetune_retrieval.yaml \
-o Arch.Backbone.pretrained=True
-o use_gpu=True
```
......@@ -57,7 +57,7 @@ python tools/train.py \
```
python tools/train.py \
-c configs/quick_start/ResNet50_flowers_retrieval_finetune.yaml \
-c configs/quick_start/ResNet50_vd_finetune_retrieval.yaml \
-o checkpoints="./output/RecModel/ppcls_epoch_5" \
-o last_epoch=5 \
-o use_gpu=True
......@@ -77,11 +77,11 @@ python tools/train.py \
```bash
python tools/eval.py \
-c ./configs/quick_start/ResNet50_flowers_retrieval_finetune.yaml \
-c ./configs/quick_start/ResNet50_vd_finetune_retrieval.yaml \
-o pretrained_model="./output/RecModel/best_model"\
```
上述命令将使用`./configs/quick_start/ResNet50_flowers_retrieval_finetune.yaml`作为配置文件,对上述训练得到的模型`./output/RecModel/best_model`进行评估。你也可以通过更改配置文件中的参数来设置评估,也可以通过`-o`参数更新配置,如上所示。
上述命令将使用`./configs/quick_start/ResNet50_vd_finetune_retrieval.yaml`作为配置文件,对上述训练得到的模型`./output/RecModel/best_model`进行评估。你也可以通过更改配置文件中的参数来设置评估,也可以通过`-o`参数更新配置,如上所示。
<a name="2"></a>
## 2. 基于Linux+GPU的模型训练与评估
......@@ -100,7 +100,7 @@ export CUDA_VISIBLE_DEVICES=0,1,2,3
python -m paddle.distributed.launch \
--gpus="0,1,2,3" \
tools/train.py \
-c ./configs/quick_start/ResNet50_flowers_retrieval_finetune.yaml
-c ./configs/quick_start/ResNet50_vd_finetune_retrieval.yaml
```
其中,`-c`用于指定配置文件的路径,可通过配置文件修改相关训练配置信息,也可以通过添加`-o`参数来更新配置:
......@@ -109,7 +109,7 @@ python -m paddle.distributed.launch \
python -m paddle.distributed.launch \
--gpus="0,1,2,3" \
tools/train.py \
-c ./configs/quick_start/ResNet50_flowers_retrieval_finetune.yaml \
-c ./configs/quick_start/ResNet50_vd_finetune_retrieval.yaml \
-o pretrained_model="" \
-o use_gpu=True
```
......@@ -127,7 +127,7 @@ export CUDA_VISIBLE_DEVICES=0,1,2,3
python -m paddle.distributed.launch \
--gpus="0,1,2,3" \
tools/train.py \
-c ./configs/quick_start/ResNet50_flowers_retrieval_finetune.yaml \
-c ./configs/quick_start/ResNet50_vd_finetune_retrieval.yaml \
-o Arch.Backbone.pretrained=True
```
......@@ -143,7 +143,7 @@ export CUDA_VISIBLE_DEVICES=0,1,2,3
python -m paddle.distributed.launch \
--gpus="0,1,2,3" \
tools/train.py \
-c ./configs/quick_start/ResNet50_flowers_retrieval_finetune.yaml \
-c ./configs/quick_start/ResNet50_vd_finetune_retrieval.yaml \
-o checkpoints="./output/RecModel/ppcls_epoch_5" \
-o last_epoch=5 \
-o use_gpu=True
......@@ -158,7 +158,7 @@ python -m paddle.distributed.launch \
```bash
python tools/eval.py \
-c ./configs/quick_start/ResNet50_flowers_retrieval_finetune.yaml \
-c ./configs/quick_start/ResNet50_vd_finetune_retrieval.yaml \
-o pretrained_model="./output/RecModel/best_model"\
```
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册