Where is new repo of tf_to_tr_image_classification for JetPack 3.3 / TensorRT 4

AI & Data Science Deep Learning (Training & Inference) TensorRT
1

In response the topic “error with cmake tf_to_trt_image_classification (jetson tx2, ubuntu 16.04, cuda 9.0 , cudnn -7.1.5)” the moderator NVES stated “tf_to_trt_image_classification repo uses TFTRT (not UFF) to convert a TF graph to TRT.”

Is there a new repo for JetPack 3.3 / TensorRT 4?
The version I used with JetPack 3.3 / TensorRT 3 used UFF in frozenToPlan.

The only version I find on GitHub, “https://github.com/NVIDIA-Jetson/tf_to_trt_image_classification”, clearly states it is for 3.2.

Is there a new version?
If so where is it?

2

Hello dbusby,

can you modify the instructions in the repo (specifically step4) to pull TensorRT 4 instead of 3?

something like
tar -xzf TensorRT-4.0.1.6.Ubuntu-16.04.4.x86_64-gnu.cuda-9.0.cudnn7.1.tar.gz

instead of

tar -xzf TensorRT-3.0.4.Ubuntu-16.04.3.x86_64.cuda-9.0.cudnn7.0.tar.gz

Then pip install the appropriate TensorRT-4 whl.

Let me know if this was your concern about the tf_to_trt_image repo not having TRT4 instructions?

3

Hello NVES,

I am using a TX2.
I most recently flashed it using JetPack 3.3.

My concerns about the tf_to_trt_image_classification repo:

1 - Is there a new repo for 3.3 that is different from 3.2?
2 - If so, where is it located?
3 - Is it loaded as part of the flash?
4 - Is it in a whl?
5 - Is it on GitHub?

Previously I used JetPack 3.2.
I was able to convert a retrained model to a plan using frozenToPlan.
Then that plan could be used by examples/classify_image/classify_image.cu to classify images using TensorRT 3.

I am trying to duplicate that for TensorRT 4

4

Hello,

As described earlier in the thread, you can modify the instructions in the repo (specifically step4) to pull TensorRT 4 instead of 3. Then, you should be able to convert your model to plan and use TRT4.

Are you running into a specific error when using TRT4?

For questions regarding the Jetson TF_to_TRT repo , please post on Jetson devtalk. (https://devtalk.nvidia.com/default/board/139/jetson-embedded-systems/)