Closed wangzhenlin123 closed 3 years ago
Hello @wangzhenlin123 , tensor support 8d tensor at most. The n-d tensor are the same, and are you asking for 3d conv? You can use addConvolutionNd
to add it. Sorry there is no existing sample that I am aware.
Forgive me for my unclear explanation,for example:“cudaMemcpyAsync(buffers[inputIndex], input, batchSize INPUT_SIZE sizeof(float), cudaMemcpyHostToDevice, stream)” We can use OpenCV to read a 2D Mat graph and put it in the Buffer, but, In the case of 3D, is it possible to build the 3D Tensorr input by reading multiple 2D images in succession and then placing them in the Buffer one by one?3D is very important to me because I work with medical 3D images .
Hello @wangzhenlin123 , yes it is the same, we can first check the input layout of your network, allocate GPU buffer has the same size, and finally use cudaMemcpyAsync to initialize the content of the GPU buffer. Did you hit any issue in doing so? thanks
I haven't tried yet,just Do some technical advice first。
Thanks @wangzhenlin123 , closing, please reopen if you have more issues, thanks
I try to inferent 3dcnn model with tensorrt,but i have no idea adout how to build 3d tensor in tensorrt with c++, could you provide some example to it?