YOLO_V8分割

YOLO_V8分割

YOLO安装

pip install ultralytics

YOLO的数据集转化看csdn

数据标注EIseg

1. EIseg这块，正常安装就好，但是numpy和各类包都容易有冲突，python版本装第一点

2. 数据标注过程中，记得把JSON和COCO都点上，把自动保存点上，如果标注后需要修改，记得手动点右下角保存，否则“空格”没有存上。

3. 预训练模型在/home/gsh/Projects/Yanbao/dataset/static_hrnet18_ocr64_cocolvis.zip

格式转化

labelme转COCO

1. paddle标注得到的是labelme格式文件，首先由于是多人标注，其中label和imagePath是不一样的，所以需要统一，该格式采用read_json_save_all.py脚本修改。在/home/gsh/Projects/Yanbao/dataset/data_for_train_2/read_json_save_all.py该脚本生成的是每个图片对应的labelme.json文件。需要输入所有标注的labelme文件夹路径，输出保存路径，和图片路径。

2. labelme转COCO采用的是官方脚本：labelme-json2labelme-coco.py 在~/Projects/PaddleSeg/EISeg/tool下

python labelme-json2labelme-coco.py [path to json dir] [path to 输出/output] --labels [path to labels.txt]
# 注意，上述的json dir要用统一后的json文件
# 例如：
python /home/gsh/Projects/PaddleSeg/EISeg/tool/labelme-json2labelme-coco.py /home/gsh/Projects/Yanbao/dataset/data_for_train_2/all_data1/tongyi_data /home/gsh/Projects/Yanbao/dataset/output --labels /home/gsh/Projects/Yanbao/dataset/data_for_train_2/all_data1/together/label/labelme/labels.txt

结果如下：

├── annotations.json
├── JPEGImages
└── Visualization

COCO转YOLOV8

1. 将上面生成的annotations.json进行处理，将JPEGImages替换为.，建议方法是VSCode打开后进行全局查找替换。
   
2. JSON2YOLO库中
   general_json2yolo.py 在~/Projects/JSON2YOLO下
   修改一下这里的路径，注意，这里要将上面的annotations.json放到一个文件夹中，然后将文件夹路径输入
   cls91to80也改成false
   
   把这里的-1去掉，这样类编号就从0开始了
   

默认生成的路径是执行代码的路径下，生成new_dir文件夹

也可以改路径，在代码的第258行

生成的是这样的
├── images
└── labels
└── annotations

调整为这样
├── images
│ ├── train
│ └── val
└── labels
├── train
└── val

至此数据处理全部结束

YOLO-seg的训练

1. 代码在/home/gsh/Projects/YOLOV8/segtrain.py，其中model = YOLO("/home/gsh/Downloads/yolov8l-seg.pt")这里面的名字写的谁，要么读取本地的，要么直接从网上下，写什么下什么

2. 记得修改yaml中的文件夹路径和类别，代码在/home/gsh/Projects/ultralytics/ultralytics/cfg/datasets/coco8-seg.yaml

3. train的选择参数很多，参考这里

   results = model.train(data="/home/gsh/Projects/ultralytics/ultralytics/cfg/datasets/coco8-seg.yaml", epochs=1000,imgsz=640, batch=-1)
   

   整体例子：

   # Load a model
   # model = YOLO("yolov8l-seg.yaml")  # build a new model from YAML
   model = YOLO("/home/gsh/Downloads/yolov8s-seg.pt")  # load a pretrained model (recommended for training)
   # model = YOLO("/home/gsh/Downloads/yolov8l-seg.pt")  # load a pretrained model (recommended for training)
   # model = YOLO("yolov8l-seg.yaml").load("/home/gsh/Downloads/yolov8l-seg.pt")  # build from YAML and transfer weights
   # model = YOLO("yolov8l-seg.yaml").load("/home/gsh/Downloads/yolov8l-seg.pt")  # build from YAML and transfer weights
   # Train the model
   results = model.train(data="/home/gsh/Projects/ultralytics/ultralytics/cfg/datasets/coco8-seg.yaml", epochs=900,imgsz=640 ,batch=32)
   

YOLO-seg的预测

predict

from ultralytics import YOLO# Load a pretrained YOLOv8n model
model = YOLO("yolov8n.pt")# Run inference on an image
results = model("bus.jpg")  # results list# View results
for r in results:print(r.boxes)  # print the Boxes object containing the detection bounding boxes

YOLO-Seg部署模型转化

onnx的部署

X64平台

模型转化代码

from ultralytics import YOLOmodel=YOLO("/home/gsh/Projects/YOLOV8/runs/segment/train8/weights/best.pt")
model.export(format="onnx", device=0, int8=True)

报错

Loading /home/gsh/Projects/YOLOV8/runs/segment/train8/weights/best.onnx for ONNX Runtime inference…
[1;31m2024-08-26 09:30:46.347354781 [E:onnxruntime:Default, provider_bridge_ort.cc:1992 TryGetProviderInfo_CUDA] /onnxruntime_src/onnxruntime/core/session/provider_bridge_ort.cc:1637 onnxruntime::Provider& onnxruntime::ProviderLibrary::Get() [ONNXRuntimeError] : 1 : FAIL : Failed to load library libonnxruntime_providers_cuda.so with error: libcudnn.so.9: cannot open shared object file: No such file or directory
[m
[0;93m2024-08-26 09:30:46.347381605 [W:onnxruntime:Default, onnxruntime_pybind_state.cc:965 CreateExecutionProviderInstance] Failed to create CUDAExecutionProvider. Require cuDNN 9.* and CUDA 12.*. Please install all dependencies as mentioned in the GPU requirements page (https://onnxruntime.ai/docs/execution-providers/CUDA-ExecutionProvider.html#requirements), make sure they’re in the PATH, and that your GPU is supported.[m

推理

直接替换掉模型路径即可

Jetson AGX

改名字

best.pt改成best-seg.pt（我也不知道为什么要加seg）据说是它模型转化要认这是个什么任务

模型转化代码

from ultralytics import YOLOmodel=YOLO("/home/gsh/Projects/YOLOV8/runs/segment/train8/weights/best.pt")
model.export(format="onnx", imgsz=(480,640), device=0, int8=True, simplify=True, task = "segment")

Tensort的部署

官网介绍

模型转化代码

from ultralytics import YOLOmodel = YOLO("yolov8n.pt")
model.export(format="engine",dynamic=True,  batch=1,  workspace=10,  int8=True,data="coco.yaml",  # 这个yaml文件要用自己训练的yaml文件
)# Load the exported TensorRT INT8 model
model = YOLO("yolov8n.engine", task="detect")# Run inference
result = model.predict("https://ultralytics.com/images/bus.jpg")

推理

直接替换掉模型路径即可