This is a hands-on, guided project on optimizing your TensorFlow models for inference with NVIDIA's TensorRT. By the end of this 1.5 hour long project, you will be able to optimize Tensorflow models using the TensorFlow integration of NVIDIA's TensorRT (TF-TRT), use TF-TRT to optimize several deep learning models at FP32, FP16, and INT8 precision, and observe how tuning TF-TRT parameters affects performance and inference throughput. Prerequisites: In order to successfully complete this project, you should be competent in Python programming, understand deep learning and what inference is, and have experience building deep learning models in TensorFlow and its Keras API. Note: This course works best for learners who are based in the North America region. We’re currently working on providing the same experience in other regions.
The first to introduce such a rare and important topic.
Great workshop, all the concepts were very well explained.
Showing: 20 of 20
Very nice project. The only note is that the installation of TensorRT is outdated. Just replace 'Install TensorFlow-GPU 2.0 and TensorRT Runtime:' section with the following. It may take some time to run.
%%bash
sudo apt install python3-libnvinfer
python3 -m pip install --upgrade tensorrt
The installation may be also verified by:
import tensorrt
print(tensorrt.__version__)
This is something that I was looking for. I've studied a lot of theories about TensorRT but this project gives a clear view of how to do it. Good job, and thanks for the awesome course.
One last thing, Please upload the TensorRT deployment of TensorFlow object detection on Jetson devices. That would be helpful
good content, but some code is out of date, especially the package installation part.
It's a very cool course, but it's outdated and underlying working environment won't let you proceed with practise just at the middle.
I do not recommend taking this type of course, take one and pass it, however after a few days I have tried to review the material, and my surprise is that it asks me to pay again to be able to review the material. Of course coursera gives me a small discount for having already paid it previously. It is very easy to download the videos and difficult to get hold of the material, but with ingenuity it is possible. Then I recommend uploading them to YouTube and keeping them private for when they want to consult (they avoid legal problems and can share with friends), then they can request a refund.
A lot has changed, float32 and float16 are now not faster than normal tensorflow. But but it is really good pipeline for model optimizing.
Great workshop, all the concepts were very well explained.
The first to introduce such a rare and important topic.
Excelent and compresed way of explaining TensorRT
Awesome project. Thank you so much.
Excellent guided course
Informative course
Very good...
The library versions is dated so it would be very hard for people, especially new learners, to setup the environment. But the lecturer and content is indeed amazing. I learned a lot about TF-TRT systematically.
Need more theoretical explanation on concepts
excellent
I expect for Coding and Implementation part but this is just theory based evaluation
The lab code is outdated and does not run. Looking at other reviews, I see that it has been an issues for quite a while already. No attempts have been made to fix it. fossils belong to a museum. Moreover, the quality of the video is extremely poor (blurred). merely watching the course (without doing it that is impossible as described above) is difficult. this course should be taken down. it steals time from the learners, which is disrespectful.
Not working now due to repo issues
Not working