widonws下yolov8模型生成wts转engine文件报错

[XuJWood]

Env

• GPU, e.g. 4070
• OS, e.g. Win10, etc.
• Cuda v11.8
• TensorRT 8.6.1

About this repo

• which branch/tag/commit are you using?
• which model? yolov5, retinaface?

yolov8-seg

Your problem

• what is your command? e.g. sudo ./yolov5 -s
• what's your output?
• what output do you expect?

我从官网下载的yolov8s-seg.pt,在tensorrt项目中利用get_wts.py生成yolov8s-seg.wts，然后在项目中生成engine文件时报错。 Loading weights: ../weights/yolov8s-seg.wts [05/27/2024-11:09:22] [W] [TRT] The implicit batch dimension mode has been deprecated. Please create the network with NetworkDefinitionCreationFlag::kEXPLICIT_BATCH flag whenever possible. [05/27/2024-11:09:22] [E] [TRT] 3: (Unnamed Layer 0) [Convolution]:kernel weights has count 864 but 0 was expected [05/27/2024-11:09:22] [E] [TRT] 4: (Unnamed Layer 0) [Convolution]: count of 864 weights in kernel, but kernel dimensions (3,3) with 3 input channels, 0 output channels and 1 groups were specified. Expected Weights count is 3 33 0 / 1 = 0 [05/27/2024-11:09:22] [E] [TRT] 4: [convolutionNode.cpp::nvinfer1::builder::ConvolutionNode::computeOutputExtents::58] Error Code 4: Internal Error ((Unnamed Layer 0) [Convolution]: number of kernel weights does not match tensor dimensions) [05/27/2024-11:09:22] [E] [TRT] 3: [network.cpp::nvinfer1::Network::addResize::1421] Error Code 3: API Usage Error (Parameter check failed at: network.cpp::nvinfer1::Network::addResize::1421, condition: input.getDimensions().nbDims > 0

请问是什么原因，应该如何解决呢

[wang-xinyu]

@lindsayshuo pls advice.

[XuJWood]

已经解决，是因为模型转化的时候 float gd = 0.33, gw = 0.50; int max_channels = 1024;没有修改，设置为0，所以通道不匹配，谢谢！

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[ranranmen]

您好，我在得到yolov8n.exe文件之后转化engine模型就遇到了下面这个问题 cmd之后 D:\YOLOv8\yolov8_tensorrt-main\build\Release>yolov8.exe -s D:\YOLOv8\weights\yolov8n.wts D:\YOLOv8\weights\yolov8n.engine s Load engine file failed: D:/YOLOv8/yolov8_tensorrt-main/weights/yolov8n.engine 很纳闷，我的环境是 cuda12.0 gtx050 TensorRT-8.6.1.6 opencv4.8.0（编译过支持cuda加速的），我现在思考是不是因为，我安装了cuda之后再装的visual studio导致的（针对这个编译前我已经把cuda相关文件复制到vs里面了），但您上面说的转化模型修改参数，我没有做过更改，直接默认的