⚡️ optim: 优化脚本

.gitignore

@@ -1,9 +1,13 @@
+# Ignore custom configuration files
+scripts/*.custom.sh
+
 tmp/
 tmps/
 temp/
 temps/
 resource/
 
+
 # Vuepress/Vitepress
 node_modules
 docs/.vuepress/.temp

README.md

@@ -1,3 +1,10 @@
 # deep-object-detect-track
 
-查看[文档](https://henryzhuhr.github.io/deep-object-detect-track/)
+查看[文档](https://henryzhuhr.github.io/deep-object-detect-track/)
+
+也可以本地启动文档：
+
+```bash
+pnpm install
+pnpm docs:dev
+```

dlinfer/detector/b_onnx.py

@@ -10,22 +10,21 @@
 
 class ONNXDetectorBackend(IDetectoBackends):
     NAME = "ONNX"
-    SUPPORTED_VERISONS = ["1.8.0"]
+    SUPPORTED_VERISONS = []
     SUPPORTED_DEVICES = available_providers
 
     def __init__(
         self,
-        device: str = "CPUExecutionProvider",
+        device: List[str] = ["CPUExecutionProvider"],
         inputs: List[str] = ["input"],  # TODO in case of multiple inputs
         outputs: List[str] = ["output"],  # TODO in case of multiple outputs
     ) -> None:
-        if device.lower() == "cpu":
-            device = "CPUExecutionProvider"
-        assert device in self.SUPPORTED_DEVICES, (
-            f"specify device {device} is not supported, "
-            f"please choose one of supported device: {self.SUPPORTED_DEVICES}"
-        )
-        self.providers = [device]
+        for provider in device:
+            assert provider in self.SUPPORTED_DEVICES, (
+                f"specify device {device} is not supported, "
+                f"please choose one of supported device: {self.SUPPORTED_DEVICES}"
+            )
+        self.providers = device
         self.ort_session: ort.InferenceSession = None
         self.inputs = inputs
         self.outputs = outputs
@@ -45,15 +44,15 @@ def load_model(self, model_path: str, verbose: bool = False) -> None:
         if verbose:
             # fmt: off
             print(self.ColorStr.info("Parsing ONNX info:"))
-            print(self.ColorStr.info("  - providers:"), self.ort_session.get_providers())
-            print(self.ColorStr.info("  --- inputs:"), binding__input_names)
-            print(self.ColorStr.info("       -- names:"), binding__input_names)
-            print(self.ColorStr.info("       - shapes:"), binding__input_shapes)
-            print(self.ColorStr.info("       -- types:"), binding__input_types)
-            print(self.ColorStr.info("  --- outputs:"), binding_output_names)
-            print(self.ColorStr.info("       -- names:"), binding_output_shapes)
-            print(self.ColorStr.info("       - shapes:"), binding_output_shapes)
-            print(self.ColorStr.info("       -- types:"), binding_output_types)
+            print(self.ColorStr.info("- providers:"), self.ort_session.get_providers())
+            print(self.ColorStr.info("--  inputs:"), binding__input_names)
+            print(self.ColorStr.info("    -  names:  "), binding__input_names)
+            print(self.ColorStr.info("    - shapes:  "), binding__input_shapes)
+            print(self.ColorStr.info("    -  types:  "), binding__input_types)
+            print(self.ColorStr.info("-- outputs:"), binding_output_names)
+            print(self.ColorStr.info("    -  names:  "), binding_output_shapes)
+            print(self.ColorStr.info("    - shapes:  "), binding_output_shapes)
+            print(self.ColorStr.info("    -  types:  "), binding_output_types)
             # fmt: on
 
         # fmt: off

dlinfer/detector/b_openvino.py

@@ -76,6 +76,7 @@ def infer(self, input: np.ndarray) -> np.ndarray:
         preds = next(iter(results.values()))
         return preds
 
-    def query_device(self):
+    @staticmethod
+    def query_device():
         """Query available devices for OpenVINO backend."""
-        return ["AUTO"] + self.core.available_devices
+        return Core().available_devices

dlinfer/detector/interface.py

@@ -48,7 +48,7 @@ def _check_version(self, version: str):
         for sv in self.SUPPORTED_VERISONS:
             if version.startswith(sv):
                 return
-        raise RuntimeError(
+        warnings.warn(
             f"{self.NAME} version {version} is not supported, "
             f"please upgrade to support version: {self.SUPPORTED_VERISONS}"
         )

docs/dataset.md

@@ -11,30 +11,31 @@ outline: deep
 
 将数据集放入如下目录
 
-```bash
+```shell
 DATASET_DIR=/path/to/dataset
 ```
 
 > 需要注意的是，数据集通常需要放置在项目外的路径，例如 `~/data` （推荐）。如果放置在项目内，导致编辑器对于项目的索引过大，会导致编辑器卡顿
 
 这里准备好了一个示例数据集，可以下载
 
-```bash
+```shell
 wget -P ~/data https://github.com/HenryZhuHR/deep-object-detect-track/releases/download/v1.0.0/bottle.tar.xz
 tar -xvf ~/data/bottle.tar.xz -C ~/data
+mv ~/data/bottle ~/data/yolodataset
 ```
 
 随后可以设置数据集目录为
-```bash
-DATASET_DIR=~/data/bottle
+```shell
+DATASET_DIR=~/data/yolodataset
 ```
 
 参考该目录构建自己的数据集，并且完成标注
 
 考虑到单张图像中可能出现不同类别的目标，因此数据集不一定需要按照类别进行划分，可以自定义划分，按照项目的需求任意归档数据集，但是请确保，每一张图像同级目录下有同名的**标签文件**
 
 按照类别划分的目录结构参考
-```bash
+```shell
 ·
 └── /path/to/dataset
     ├── class_A         
@@ -48,7 +49,7 @@ DATASET_DIR=~/data/bottle
 ```
 
 不进行类别划分的目录结构参考
-```bash
+```shell
 ·
 └── /path/to/dataset    
     ├─ file_1.jpg  
@@ -60,7 +61,7 @@ DATASET_DIR=~/data/bottle
 ## 启动标注工具
 
 使用 labelImg 标注，安装并启动
-```bash
+```shell
 pip install labelImg
 labelImg
 ```
@@ -101,11 +102,11 @@ labelImg
 ## 数据处理
 
 运行脚本，生成同名目录，但是会带 `-organized` 后缀，例如
-```bash
-python dataset-process.py --datadir ~/data/bottle
+```shell
+python dataset-process.py --datadir ~/data/yolodataset
 ```
 
-生成 `~/data/bottle-organized` 用于数据集训练，并且该目录为 yolov5 中指定的数据集路径
+生成的目录 `~/data/yolodataset-organized` 用于数据集训练，并且该目录为 yolov5 中指定的数据集路径
 
 如果不需要完全遍历数据集、数据集自定义路径，则在 `get_all_label_files()` 函数中传入自定义的 `custom_get_all_files` 函数，以获取全部文件路径，该自定义函数可以参考 `default_get_all_files()`
 
@@ -114,15 +115,13 @@ def default_get_all_files(directory: str):
     file_paths: List[str] = []
     for root, dirs, files in os.walk(directory):
         for file in files:
-            if file in [".DS_Store"]:
-                continue
             file_paths.append(os.path.join(root, file))
     return file_paths
 ```
 
 并且在调用的时候传入该参数
 
-```bash
+```python
 # -- get all label files, type: List[ImageLabel]
 label_file_list = get_all_label_files(args.datadir) # [!code --]
 label_file_list = get_all_label_files(          # [!code ++]

docs/deploy.md

@@ -9,41 +9,135 @@ outline: deep
 
 ## 导出模型
 
-已经编写了一个脚本，可以直接导出
+提供一个导出脚本 `scripts/train.sh`，复制一份到项目目录下进行自定义修改（推荐）
 
 ```shell
-# 查看脚本，修改参数
-bash scripts/export-yolov5.sh
+cp scripts/export-yolov5.sh scripts/export-yolov5.custom.sh
 ```
 
-也可以自行导出，进入 yolov5 项目目录
+查看脚本 `scripts/export-yolov5.custom.sh` ，根据项目需求修改参数后执行
 
 ```shell
-cd projects/yolov5
+bash scripts/export-yolov5.custom.sh
+#zsh scripts/export-yolov5.custom.sh # zsh
 ```
-设定参数，执行
-```shell
-python export.py \
-    --weights ../weights/yolov5s.pt \
-    --data data/coco128.yaml \
-    --include onnx openvino 
-```
-- `--weights` 模型路径: 修改为自己训练好的模型路径
-- `--include` 导出类型: 可以导出多个模型，用空格分隔
-- `--data` 训练模型使用的数据集，主要利用到里面的类别数据
+
 
 ## 部署模型
 
 部署模型请单独创建虚拟环境，并根据部署中提到的依赖进行**最小化安装**，而避免引入过多不必要的包 （例如 pytorch），部署环境均经过测试
 
 ### ONNX 部署
 
+修改 `scripts/variables.custom.sh` 文件中 `ENV_NAME` 如下
+
+```shell
+# export ENV_NAME="" # -- Uncomment to customize the environment name # [!code --]
+export ENV_NAME="deploy-onnx" # [!code ++]
+export ENV_PATH=$BASE_ENV_PATH/.env/$ENV_NAME
+```
+
+然后执行脚本创建虚拟环境，并激活
+
+::: code-group
+
+```shell [使用 venv]
+bash scripts/create-python-env.sh -e venv -ni
+#zsh scripts/create-python-env.sh -e venv -ni # zsh
+```
+
+```shell [使用 conda]
+bash scripts/create-python-env.sh -e conda -ni
+#zsh scripts/create-python-env.sh -e conda -ni # zsh
+```
+
+:::
+
+根据上述脚本运行输出激活环境
+
+```shell
+- [INFO] Run command below to activate the environment:
+... # 复制这里出现的激活命令并执行
+```
+
+手动安装依赖
+
+```shell
+pip install -r requirements/requirements.txt
+pip install onnxruntime # CPU 版本
+# pip install onnxruntime-gpu # GPU 版本
+```
+
+修改 `infer.py` 文件，指定模型路径
+```python
+## ------ ONNX ------
+onnx_backend = backends.ONNXBackend
+print("-- Available devices:", providers := onnx_backend.SUPPORTED_DEVICES)
+detector = onnx_backend(
+    device=providers, inputs=["images"], outputs=["output0"]
+)
+```
+
+然后执行推理脚本
+
+```shell
+python infer.py --model .cache/yolov5/yolov5s.onnx
+```
+
+
 ### OpenVINO 部署
 
+修改 `scripts/variables.custom.sh` 文件中 `ENV_NAME` 如下
+
 ```shell
-python infer.py
+# export ENV_NAME="" # -- Uncomment to customize the environment name # [!code --]
+export ENV_NAME="deploy-ov" # [!code ++]
+export ENV_PATH=$BASE_ENV_PATH/.env/$ENV_NAME
+```
+
+然后执行脚本创建虚拟环境，并激活
+
+::: code-group
+
+```shell [使用 venv]
+bash scripts/create-python-env.sh -e venv -ni
+#zsh scripts/create-python-env.sh -e venv -ni # zsh
+```
+
+```shell [使用 conda]
+bash scripts/create-python-env.sh -e conda -ni
+#zsh scripts/create-python-env.sh -e conda -ni # zsh
 ```
 
+:::
+
+根据上述脚本运行输出激活环境
+
+```shell
+- [INFO] Run command below to activate the environment:
+... # 复制这里出现的激活命令并执行
+```
+
+手动安装依赖
+
+```shell
+pip install -r requirements/requirements.txt
+pip install openvino
+```
+
+修改 `infer.py` 文件，指定模型路径
+```python
+## ------ ONNX ------
+ov_backend = backends.OpenVINOBackend
+print("-- Available devices:", ov_backend.query_device())
+detector = ov_backend(device="AUTO")
+```
+
+然后执行推理脚本
+
+```shell
+python infer.py --model .cache/yolov5/yolov5s_openvino_model/yolov5s.xml
+```
 
 ### TensorRT 部署
 

docs/index.md

@@ -9,7 +9,7 @@ hero:
   actions:
     - theme: brand
       text: 项目源码
-      link: https://henryzhuhr.github.io/deep-object-detect-track/
+      link: https://github.com/HenryZhuHR/deep-object-detect-track
     - theme: alt
       text: 项目文档
       link: /install