TensorRTX反序列fp32模型量化失败，原因是因为numclasses数量不对。当前为v8s_seg分割模型，但是报错，生成的engine文件为0kb故为错误的。

光芒再现dev

已于 2024-04-10 15:52:10 修改

阅读量1.1k

点赞数 26

CC 4.0 BY-SA版权

文章标签： opencv 计算机视觉人工智能 YOLO

于 2024-04-10 15:49:17 首次发布

本文链接：https://blog.youkuaiyun.com/blink182007/article/details/137599830

先附上解决方案：参考修改config.h文件

[Convolution]:kernel weights has count 3840 but 32640 was expected，yolov5s生成engine模型失败_yolov5s.engine-优快云博客https://blog.youkuaiyun.com/pvmsmfchcs/article/details/131388796修改这个config.h文件，然后继续修改

当前分割背景的类别不要算进去，训练的时候，是几个类别，这里就是几个，不要额外加一

遇见警告如下：

[04/10/2024-15:51:05] [W] [TRT] If this is not the desired behavior, please modify the weights or retrain with regularization to reduce the magnitude of the weights.
[04/10/2024-15:51:06] [W] [TRT] Weights [name=(Unnamed Layer* 346) [Deconvolution].weight] had the following issues when converted to FP16:
[04/10/2024-15:51:06] [W] [TRT] - Subnormal FP16 values detected.
[04/10/2024-15:51:06] [W] [TRT] If this is not the desired behavior, please modify the weights or retrain with regularization to reduce the magnitude of the weights.
[04/10/2024-15:51:06] [W] [TRT] Weights [name=(Unnamed Layer* 346) [Deconvolution].weight] had the following issues when converted to FP16:
[04/10/2024-15:51:06] [W] [TRT] - Subnormal FP16 values detected.
[04/10/2024-15:51:06] [W] [TRT] If this is not the desired behavior, please modify the weights or retrain with regularization to reduce the magnitude of the weights.
[04/10/2024-15:51:06] [W] [TRT] Weights [name=(Unnamed Layer* 346) [Deconvolution].weight] had the following issues when converted to FP16:
[04/10/2024-15:51:06] [W] [TRT] - Subnormal FP16 values detected.
[04/10/2024-15:51:06] [W] [TRT] If this is not the desired behavior, please modify the weights or retrain with regularization to reduce the magnitude of the weights.
[04/10/2024-15:51:06] [W] [TRT] Weights [name=(Unnamed Layer* 346) [Deconvolution].weight] had the following issues when converted to FP16:
[04/10/2024-15:51:06] [W] [TRT] - Subnormal FP16 values detected.
[04/10/2024-15:51:06] [W] [TRT] If this is not the desired behavior, please modify the weights or retrain with regularization to reduce the magnitude of the weights.
[04/10/2024-15:51:06] [W] [TRT] Weights [name=(Unnamed Layer* 346) [Deconvolution].weight] had the following issues when converted to FP16:
[04/10/2024-15:51:06] [W] [TRT] - Subnormal FP16 values detected.
[04/10/2024-15:51:06] [W] [TRT] If this is not the desired behavior, please modify the weights or retrain with regularization to reduce the magnitude of the weights.
[04/10/2024-15:51:06] [W] [TRT] Weights [name=(Unnamed Layer* 346) [Deconvolution].weight] had the following issues when converted to FP16:
[04/10/2024-15:51:06] [W] [TRT] - Subnormal FP16 values detected.
[04/10/2024-15:51:06] [W] [TRT] If this is not the desired behavior, please modify the weights or retrain with regularization to reduce the magnitude of the weights.
[04/10/2024-15:51:06] [W] [TRT] Weights [name=(Unnamed Layer* 346) [Deconvolution].weight] had the following issues when converted to FP16:
[04/10/2024-15:51:06] [W] [TRT] - Subnormal FP16 values detected.
[04/10/2024-15:51:06] [W] [TRT] If this is not the desired behavior, please modify the weights or retrain with regularization to reduce the magnitude of the weights.