fix bug dataset.py

2021-05-31 14:13:12 +08:00 · 2021-05-31 14:13:12 +08:00 · 10714cde2c
parent 61c96a444f
commit 10714cde2c
4 changed files with 24 additions and 63 deletions
--- a/model_zoo/official/cv/shufflenetv1/README_CN.md
+++ b/model_zoo/official/cv/shufflenetv1/README_CN.md
@ -14,10 +14,6 @@
    - [评估过程](#评估过程)
        - [启动](#启动-1)
        - [结果](#结果-1)
-    - [导出过程](#导出过程)
-        - [导出](#导出)
-    - [推理过程](#推理过程)
-        - [推理](#推理)
 - [模型说明](#模型说明)
    - [训练性能](#训练性能)
 - [随机情况的描述](#随机情况的描述)
@ -64,7 +60,6 @@ ShuffleNetV1的核心部分被分成三个阶段，每个阶段重复堆积了
    ├─run_standalone_train.sh                 # Ascend环境下的单卡训练脚本
    ├─run_distribute_train.sh                 # Ascend环境下的八卡并行训练脚本
    ├─run_eval.sh                             # Ascend环境下的评估脚本
-    ├─run_infer_310.sh                        # Ascend 310 推理shell脚本
  ├─src
    ├─dataset.py                              # 数据预处理
    ├─shufflenetv1.py                         # 网络模型定义
@ -78,7 +73,6 @@ ShuffleNetV1的核心部分被分成三个阶段，每个阶段重复堆积了
  ├─default_config.yaml                       # 参数文件
  ├─train.py                                  # 网络训练脚本
  ├─export.py                                 # 模型格式转换脚本
-  ├─postprogress.py                           # 310推理后处理脚本
  └─eval.py                                   # 网络评估脚本
  └─mindspore_hub_conf.py                     # hub配置脚本
 ```
@ -182,6 +176,7 @@ result:{'Loss': 2.0479587888106323, 'Top_1_Acc': 0.7385817307692307, 'Top_5_Acc'

 # (1) 选择a(修改yaml文件参数)或者b(ModelArts创建训练作业修改参数)其中一种方式。
 #       a. 设置 "enable_modelarts=True" 。
+#          设置 "is_distributed=True"
 #          设置 "save_ckpt_path=/cache/train/outputs_imagenet/"
 #          设置 "train_dataset_path=/cache/data/train/train_dataset/"
 #          设置 "resume=/cache/data/train/train_predtrained/pred file name" 如果没有预训练权重 resume=""
@ -216,36 +211,6 @@ result:{'Loss': 2.0479587888106323, 'Top_1_Acc': 0.7385817307692307, 'Top_5_Acc'
 # (7) 开始模型的推理。
 ```

-## 导出过程
-
-### 导出
-
-```shell
-python export.py --ckpt_file [CKPT_PATH] --device_target [DEVICE_TARGET] --file_format [EXPORT_FORMAT] --batch_size [BATCH_SIZE]
-```
-
-`EXPORT_FORMAT` 可选 ["AIR", "MINDIR"]
-
-## 推理过程
-
-### 推理
-
-在推理之前需要先导出模型，AIR模型只能在昇腾910环境上导出，MINDIR可以在任意环境上导出。
-
-```shell
-# 昇腾310 推理
-bash run_infer_310.sh [MINDIR_PATH] [DATA_PATH] [LABEL_FILE] [DEVICE_ID]
-```
-
-注: Densnet121网络使用ImageNet数据集,图片的label是将文件夹排序后从0开始编号所得的数字.
-
-推理的结果保存在当前目录下，在acc.log日志文件中可以找到类似以下的结果。
-Densenet121网络使用ImageNet推理得到的结果如下:
-
-  ```log
-  Top_1_Acc=73.85%, Top_5_Acc=91.526%
-  ```
-
 # 模型说明

 ## 训练性能
--- a/model_zoo/official/cv/shufflenetv1/default_config.yaml
+++ b/model_zoo/official/cv/shufflenetv1/default_config.yaml
@ -23,12 +23,12 @@ device_id: 0
 # Training options
 epoch_size: 250
 keep_checkpoint_max: 5
-save_ckpt_path: "/data1/mjq/ckpt/shufflenetv1/test_before_pr/"
+save_ckpt_path: "./"
 save_checkpoint_epochs: 1
 save_checkpoint: True
 amp_level: "O3"
 is_distributed: False
-train_dataset_path: "/data1/mjq/dataset/ImageNet_Original/train/"
+train_dataset_path: ""
 resume: ""

 # Dataset config
@ -49,10 +49,15 @@ momentum: 0.9

 # ======================================================================================
 # Eval options
-ckpt_path: "/data1/mjq/ckpt/shufflenetv1/shufflenetv1_1-250_1251.ckpt"
-eval_dataset_path: "/data1/mjq/dataset/ImageNet_Original/validation_preprocess/"
+ckpt_path: ""
+eval_dataset_path: ""


+# ======================================================================================
+# export options
+file_name: "shufflenetv1"
+file_format: "MINDIR"
+
 ---
 # Help description for each configuration
 enable_modelarts: "Whether training on modelarts default: False"
@ -65,3 +70,6 @@ enable_profiling: "Whether enable profiling while training default: False"
 is_distributed: "distributed training"
 resume: "resume training with existed checkpoint"
 model_size: "shuffleNetV1 model size choices 2.0x, 1.5x, 1.0x, 0.5x"
+device_id: "device id"
+file_name: "output file name"
+file_format: "file format choices [AIR MINDIR ONNX]"
--- a/model_zoo/official/cv/shufflenetv1/export.py
+++ b/model_zoo/official/cv/shufflenetv1/export.py
@ -14,40 +14,28 @@
 # ============================================================================
 """
 ##############export checkpoint file into air, onnx, mindir models#################
-python export.py
+suggest run as python export.py --file_name [file name] --ckpt_path [ckpt path] --file_format [file format]
 """
-import argparse
-import numpy as np

+import numpy as np
 import mindspore as ms
 from mindspore import Tensor, load_checkpoint, load_param_into_net, export, context
-
+from src.model_utils.config import config
 from src.shufflenetv1 import ShuffleNetV1

-parser = argparse.ArgumentParser(description='ShuffleNetV1 export')
-parser.add_argument("--device_id", type=int, default=0, help="device id")
-parser.add_argument("--batch_size", type=int, default=1, help="batch size")
-parser.add_argument("--ckpt_file", type=str, required=True, help="checkpoint file path.")
-parser.add_argument("--file_name", type=str, default="shufflenetv1", help="output file name.")
-parser.add_argument('--file_format', type=str, choices=["AIR", "ONNX", "MINDIR"], default='AIR', help='file format')
-parser.add_argument("--device_target", type=str, choices=["Ascend", "GPU", "CPU"], default="Ascend",
-                    help="device target")
-parser.add_argument('--model_size', type=str, default='2.0x', choices=['2.0x', '1.5x', '1.0x', '0.5x'],
-                    help='shufflenetv1 model size')

-args = parser.parse_args()
+context.set_context(mode=context.GRAPH_MODE, device_target=config.device_target)
+if config.device_target == "Ascend":
+    context.set_context(device_id=config.device_id)

-context.set_context(mode=context.GRAPH_MODE, device_target=args.device_target)
-if args.device_target == "Ascend":
-    context.set_context(device_id=args.device_id)

 if __name__ == '__main__':

-    net = ShuffleNetV1(model_size=args.model_size)
+    net = ShuffleNetV1(model_size=config.model_size)

-    param_dict = load_checkpoint(args.ckpt_file)
+    param_dict = load_checkpoint(config.ckpt_path)
    load_param_into_net(net, param_dict)

    image_height, image_width = (224, 224)
-    input_arr = Tensor(np.ones([args.batch_size, 3, image_height, image_width]), ms.float32)
-    export(net, input_arr, file_name=args.file_name, file_format=args.file_format)
+    input_arr = Tensor(np.ones([config.batch_size, 3, image_height, image_width]), ms.float32)
+    export(net, input_arr, file_name=config.file_name, file_format=config.file_format)
--- a/model_zoo/official/cv/shufflenetv1/src/dataset.py
+++ b/model_zoo/official/cv/shufflenetv1/src/dataset.py
@ -13,7 +13,7 @@
 # limitations under the License.
 # ============================================================================
 """Data operations, will be used in train.py and eval.py"""
-from src.config import config
+from src.model_utils.config import config
 import mindspore.common.dtype as mstype
 import mindspore.dataset as ds
 import mindspore.dataset.transforms.c_transforms as C2