convert ssd_resnet to pure TRT

Hi,

I’m strugggeling to convert a SSD_RESNET50 Tensorflow frozen graph to UFF and finally to pure TRT binary format.

This is basically because of unsupported nodes (‘slice’ in this case but there will be probably further ones I have not reached on the conversion yet).

I followed the excellent thread for converting SSD_MOBILENET_V2 (https://devtalk.nvidia.com/default/topic/1049802/jetson-nano/object-detection-with-mobilenet-ssd-slower-than-mentioned-speed/post/5397858/), however there are still untranslatable nodes.

Has anyone created the needed custom plugins and would like to share?

Hi,

It’s recommended to check this GitHub first:
https://github.com/AastaNV/TRT_object_detection

Which JetPack version do you use?
Slice layer is supported in TensorRT and Uff parser after v5.1.6.1, which is JetPack 4.2.1 above.
https://docs.nvidia.com/deeplearning/sdk/tensorrt-archived/tensorrt-515/tensorrt-support-matrix/index.html

If you are using JetPack4.2.2, could you share your log with us?

Thanks.