Tx2 GIE Model Max output mem?

Robotics & Edge Computing Jetson & Embedded Systems Jetson TX2
1

I am trying to portable my caffe segmentation net to tx2, because some custom layer have learned params, It seems there is no sample about load custom layer learnable params from trained caffe model.

So I split my net, I want exectue speed part of net with tensorrt, but I find that there seem a limit about gie model output max mem, While my output need 2048 * 60 * 30 * sizeof(float) mem, and when I set out mem to this, the program crashed with segment fault, I set output mem 1/10 times as correct mem, it show nothing about crash error. But the gie model generated with wrong output mem is useless.

Can someone pls help me solve this issue?

2

Hi,

1. You can write your custom layer with TensorRT plugin API.
The updated parameter can be handled with std file read/write in the plugin class.
Here is a sample for your reference: [url]https://github.com/AastaNV/Face-Recognition[/url]

2. There is no limitation in tensorRT output amount.
Please remember to mark the layer as output to make it accessible.
[url]Face-Recognition/tensorNet.cpp at master · AastaNV/Face-Recognition · GitHub

Thanks.