OpenAskDragon
OpenAskDragon
> > > that's strange, what's your cuda version? try `nvcc --version` maybe this option does not apply to all cuda versions, mine now is 11.6.2, I updated it >...
I rebuild my pcl library with "-WITH_CUDA = TRUE "，but still meet this problem terminate called after throwing an instance of 'thrust::system::system_error' what(): for_each: failed to synchronize: cudaErrorIllegalAddress: an illegal...
Thanks. I found that my code runs successfully when I comment out the following lines: // std::cout
> > I rebuild my pcl library with "-WITH_CUDA = TRUE "，but still meet this problem terminate called after throwing an instance of 'thrust::system::system_error' what(): for_each: failed to synchronize: cudaErrorIllegalAddress:...
> target:204761[pts] source:59051[pts] --- ndt_cuda (P2D) --- terminate called after throwing an instance of 'thrust::system::system_error' what(): reduce failed to synchronize: cudaErrorIllegalAddress: an illegal memory access was encountered Aborted (core dumped)...
> zixia/wechat:3.3.0.115 is not working on my Fedora laptop. sob > > ``` > [DoChat] 盒装微信 v0.14.4 > Disabling patch for /home/user/.wine/drive_c/users/user/AppData/Roaming/Tencent/WeChat ... > Disabling patch for /home/user/.wine/drive_c/users/user/Application Data/Tencent/WeChat ......
> Bump, same error or Ubuntu 24.04 + rolling > > ``` > Python 3.11.5 (main, Sep 11 2023, 13:54:46) [GCC 11.2.0] on linux > Type "help", "copyright", "credits" or...
同问，希望林博能提供一下真值
您好，我采用了您的方案，但是我这边只能用RTX4060Laptop，我的环境为TensorRT-8.6.1.6、cuda-11.8、cudnn 8.9，最终还是不能进行superglue的匹配，但是我将Tensorrt的版本降到8.5.1.7，匹配可以正常运行，如下图所示  **我有一个问题，就是如果我改变了图片的尺寸，那么我就需要重新生成engine吗，这个生成的过程真的超级慢** 还有我的推理时间如下所示： First image feature points number: 433 First image infer cost 12.0606 MS Second image feature points number: 457 Second image infer cost 12.202 MS Match...
> Tensor8.5应该与8.6都是支持40系列的，我第一次安装的时候用的就是8.5.1.7，但是整个系统卡死了。我没有打断点来判断是哪里出问题了，直接找了8.6的tensorRT，然后把docker的cudnn换为了8.9(一开始的docker是8.6的cudnn)。至于运行时间 我感觉不太正常，跟我差的有点多(我的也够离谱的)，你是运行100次之后的结果么。或者可以加微信讨论一下(我的CSDN博客名是Super_coconut，咱私聊就可以加微信了，我也是一个初学者)。我在作者的100次的for循环里面加了一个重新读取图片的操作之后，这样每一次superpoint推理的时间就变成了2ms(很神奇，因为我没有计算读取图片的时间)。如果每一次只推理一张图片，或者只匹配一次的时间是20ms与500ms，这种情况可能是因为单次操作加载模型什么的比较废时间。 > > 这是我的执行结果，看起来匹配的结果要比你的好一点  您好，我在CSDN私聊您了