大佬，yolov5m无法运行

[rmwaring]

我使用最新的yolov5m.pt转换成wts格式后，可以成功编译，但是无法生成引擎
出现了如下错误：

Loading weights: ./yolov5m.wts
yolov5-m2: ../yolov5-m2/common.hpp:128：std::map<std::__cxx11::basic_string<char>, nvinfer1::Weights> loadWeights(std::__cxx11::string): 假设 ‘input.is_open() && "Unable to load weight file."’ 失败。
程序异常结束。

我之前使用wangxinyu的yolov5s可以成功生成引擎，应该不是环境问题吧？
我对比了我们的运行环境，是一致的。
感谢您的帮助

[AIpakchoi]

@rmwaring 这是我刚才编译生成engine的yolov5m

[AIpakchoi]

@rmwaring 最近Ultralytics 的yolov5更新到v2.0了，我没试他新上传的pt文件。我用的是7月初的pt文件转的。

[PiyalGeorge]

@AIpakchoi , Do you have that old pretrained '.pt' file for yolov5m, which works for your repo?

[AIpakchoi]

@PiyalGeorge You can download pt files of ultralytics/yolov5(version 1.0 ) on BaiduCloud,paddwd: 9cw1

[PiyalGeorge]

@AIpakchoi Thanks, By the way, Can i ask you a doubt, - Which model works best in terms of detection accuracy and fps(near-realtime-fps) in Xavier NX board(whether its yolov5s, or yolov5m, or yolov4 versions, or any other model)?

[PiyalGeorge]

@AIpakchoi , Sir can you kindly upload yolov5m model(old version) here since i'm not able to download from baidu(Even i'm not able to create a baidu account).

[AIpakchoi]

@PiyalGeorge Hi, I'm sorry that I did not test yolov5 in Xavier NX board. The pt file is so big that I can not upload it here.The following link is the old version of yolov5, but I find v2.0 model override v1.0 in google drive.
https://github.com/ultralytics/yolov5/releases/tag/v1.0
You can give me an email address and I will send the file to you.

[PiyalGeorge]

Thank you, here is my email - george6666@gmail.com

[AIpakchoi]

@PiyalGeorge Hi, I sent the Ultralytics_YOLOv5_v1.0pt.zip to your email just now.

[PiyalGeorge]

@AIpakchoi , Thanks alot, got it

[BaofengZan]

2.0已经弄好，感谢几位大佬的开源。
https://github.com/BaofengZan/yolov5_2.0-TensorRt

[AIpakchoi]

@BaofengZan haha, nice!
多谢多谢了，回头换v2.0玩玩。

[pyhcxf]

@rmwaring 你好，请问你的这个问题是怎么解决的呀？