Hello trend,
This is Satya Mallick from LearnOpenCV.com
Today's post is EPIC!
We are stepping into the world of deploying industrial-grade, production-ready computer vision applications. We use similar pipelines in our computer vision consulting company.
Specifically, today you will learn about building industrial embedded deep learning inference pipelines with TensorRT in python.
Is that a mouthful? A tad scary? Don't worry; we will break it down for you.
Beginners will find it very informative, and experts will be pleasantly surprised by the article's depth.
You will learn about TensorRT and its API concepts. You will understand optimization strategies like Reduced mixed-precision, Layer fusion, Kernel auto-tuning, Buffer reuse, and Multi-stream execution.
Next, we will show a practical example of speeding up semantic segmentation by 10x.
Finally, you will get a general understanding of the NVIDIA Jetson family of boards with their CPUs, GPUs, and Deep Learning Accelerators (DLA).
We have tested the code on NVIDIA Jetson AGX Xavier (JAX). Still, the concepts of TensorRT we will learn also apply equally to data center applications and other Jetson boards like Jetson Xavier NX. People interested in automotive applications like autonomous driving will be thrilled to know that the code in this post will also work on the NVIDIA Drive PX. With two exceptions (int8 precision and DLA inference), you will be able to run all the code on the much more affordable Jetson Nano as well.
Without further ado, let's dive into the post.
The code is at
https://github.com/spmallick/learnopencv/tree/master/industrial_cv_TensorRT_python
This four-part epic series will be your guide to TensorRT. Today we will cover Part 1, and the remaining parts will go live in the next 2 months.
- Building industrial embedded deep learning inference pipelines with TensorRT in python
- Building industrial embedded deep learning inference pipelines with TensorRT in C++
- Building industrial computer vision pipelines with Vision Programming Interface (VPI)
- Using real-time Linux kernel on embedded devices for safety-critical applications.
Blog Olympics Results
Drumroll! The LearnOpenCV Blog Olympics results are out.
https://learnopencv.com/blog-olympics/
Congratulations to all the participants and winners!
The goal of this Blog Olympics was to encourage people to start putting their thoughts in words.
Good writing is one of the most rewarding skills to have in the industry. It helps you explain ideas to others, and more importantly clarifies your own thinking. It gives you a huge edge over other engineers who do not appreciate its value.
We will work with the winners, and publish their submissions over the next few months.
Career in AI?
A few years back, in our consulting work, we used to do just a few types of projects; some companies wanted to track people, others wanted to count cars, and a few different kinds of classification tasks.
In the last four years, the variety of work has exploded. We now work on projects that involve sports analytics, document analysis in a few different languages, medical diagnostics, health/wellness, AI-assisted photography for real estate and car dealerships, safety/security, to name a few.
We have deployed solutions on the cloud, mobile devices, edge hardware, and drones.
Imagine the explosion in computer vision and AI jobs in general if a single company like ours gets to work on so many exciting projects.
Want to start a career in computer vision and AI in 2022? The Official AI Courses by OpenCV.org are the best place to start and become an expert.
| JOIN THE AI REVOLUTION |
Cheers!
Satya
Courses / Webinar / Jobs
Facebook / LinkedIn / Twitter / Instagram
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.