Tlt-converter "Unsupported operation _BatchTilePlugin_TRT" Error

[gadgadk]

**• Hardware Platform Jetson Xavier AGX
• DeepStream Version: 5.0
• TensorRT Version: 7.1

Hi,
After exporting Retinanet int8 model files (.engine .etlt .bin) using the jupyter example, i’m trying to convert the model for use with the deepstream python app “imagedata-multistream”.
Unfortunately, i’m receiving this error at the final step of the TLT guide:

chmod +x tlt-converter
export API_KEY=################################
export OUTPUT_NODES=NMS
export INPUT_DIMS=3,416,416
export D_TYPE=int8
export ENGINE_PATH=/mnt/Xavier/trt.int8.engine
export MODEL_PATH=/mnt/Xavier/retinanet_resnet_epoch_019.etlt
export CALIB=/mnt/Xavier/cal.bin

./tlt-converter -k $API_KEY -o $OUTPUT_NODES -c $CALIB -d $INPUT_DIMS -e $ENGINE_PATH $MODEL_PATH
[ERROR] UffParser: Validator error: FirstDimTile_4: Unsupported operation _BatchTilePlugin_TRT
[ERROR] Failed to parse the model, please check the encoding key to make sure it’s correct
[ERROR] Network must have at least one output
[ERROR] Network validation failed.
[ERROR] Unable to create engine
Segmentation fault (core dumped)

I’ve tried the solution offered here:

Failling in building sample from TLT-DEEPSTREAM

desktop:~/TensorRT/build/out$ ll -sh total 4.5M 4.0K drwxr-xr-x 2 dewei dewei 4.0K Jun 8 21:30 ./ 4.0K drwxr-xr-x 9 dewei dewei 4.0K Jun 8 21:29 ../ 0 lrwxrwxrwx 1 dewei dewei 26 Jun 8 21:30 libnvinfer_plugin.so -> libnvinfer_plugin.so.7.0.0* 0 lrwxrwxrwx 1 dewei dewei 28 Jun 8 21:30 libnvinfer_plugin.so.7.0.0 -> libnvinfer_plugin.so.7.0.0.1* 4.5M -rwxr-xr-x 1 dewei dewei 4.5M Jun 8 21:30 libnvinfer_plugin.so.7.0.0.1*

But i’m still getting the same error.

***I’ve downloaded the latest version from GitHub - NVIDIA/TensorRT: NVIDIA® TensorRT™ is an SDK for high-performance deep learning inference on NVIDIA GPUs. This repository contains the open source components of TensorRT. · GitHub.

***For some reason these commands don’t produce an “out” folder, but the make is still successful:

/usr/local/bin/cmake .. -DGPU_ARCHS=72 -DTRT_LIB_DIR=/usr/lib/aarch64-linux-gnu/ -DCMAKE_C_COMPILER=/usr/bin/gcc -DTRT_BIN_DIR=pwd/out
make nvinfer_plugin -j$(nproc)

Any other solutions, other than the one to do with plugin.so?

[Morganh]

Could you please find the reason why there is not an “out” folder? If possible, please share your step to build the TRT OSS libnvinfer_plugin.so.

[gadgadk]

Hi,
I manually removed the cmake folder where i cloned the git and the one from usr/bin/cmake. Started the installation from the beginning and now everything runs smoothly. As long as you know how to handle the plugin.so part properly stuff works, consider adding the instructions in the aforementioned post to the TLT guide maybe? Comparing my output to the expected output with “ll -sh /usr/lib/aarch64-linux-gnu/libnvinfer_plugin*” really helped me understand which files needed to be replaced.

Anyway you can close this thread, thanks!