Wrong Output Detection SSD inference on Jetson TX1

Autonomous Machines Jetson & Embedded Systems Jetson TX1
1

Hi,

I have fine tuned Mobile Net SSD for 2 classes. When I am using it on x86_64 the results are perfect with TensorRT. When I port this model onto Jetson TX1, all the layers are correct until createSSDDetectionOutput plugin. I have included two outputs for this layer. This works correctly on x86_64 and not on Jetson? This layer works correctly again with 20 classes. When it comes to 2 classes, this works wrong.

Thank you

2

Hi,

createSSDDetectionOutput is not a official release function and cannot be customized.
It’s recommended to implement your requirement with plugin API.

Here is a tutorial for your reference:

Thanks.