Quick Deploy: Object Detection via NGC on Vertex AI Workbench Google Cloud

📚📚 About

In this one, i’ll show you how to deploy one of many’s NVIDIA’s jupyter notebooks available on NGC, which in this case is object detection using TAO Detectnet on Vertex AI Workbench google cloud via NVIDIA’s new feature called Quick Deploy. For the development of AI applications on GPU-powered on-premise and cloud instances, I find that NGC catalog and Vertex AI Workbench google cloud are the perfect combination that maximizes or AI workflow with minimal efforts on worrying about cloud or IT infrastructure

⏲⏲Outline
00:00 Intro
00:28 NGC Catalog
01:33 NGC Models
01:57 NGC Collections
02:37 Deploy to Managed Notebook on Vertex AI Workbench Google Cloud
03:41 Preparing our Managed Notebook
04:32 Object Detection TAO DetectNet Jupyter Notebook
06:12 Set up env variables
06:48 Prepare dataset and pre-trained model
07:04 Download the dataset
07:42 Verify downloaded dataset
08:05 Verify integrity of zip files (sanity check)
08:19 Unpacking the downloaded dataset
08:30 Verify the number of images and labels in dataset
08:55 Sample KITTI labels
09:01 Prepare tensorflow records from KITTI format dataset
10:00 Download pre-trained model
10:57 Run TAO training
11:31 Evaluate the trained model
11:44 Prune the trained model
12:11 Re-train the pruned model
12:47 Evaluate the retrained model
13:24 Visualize inferences (object detection)
15:24 Outro

💰 Donate to help the channel
BTC wallet - 3KnwXkMZB4v5iMWjhf1c9B9LMTKeUQ5viP
ETH wallet - 0x44F561fE3830321833dFC93FC1B29916005bC23f
DOGE wallet - DEvDM7Pgxg6PaStTtueuzNSfpw556vXSEW
API3 wallet - 0xe447602C3073b77550C65D2372386809ff19515b
DOT wallet - 15tz1fgucf8t1hAdKpUEVy8oSR8QorAkTkDhojhACD3A4ECr
ARPA wallet - 0xf54bEe325b3653Bd5931cEc13b23D58d1dee8Dfd
QNT wallet - 0xDbfe00E5cddb72158069DFaDE8Efe2A4d737BBAC
AAVE wallet - 0xD9Db74ac7feFA7c83479E585d999E356487667c1
AGLD wallet - 0xF203e39cB3EadDfaF3d11fba6dD8597B4B3972Be
AERGO wallet - 0xd847D9a2EE4a25Ff7836eDCd77E5005cc2E76060
AST wallet - 0x296321FB0FE1A4dE9F33c5e4734a13fe437E55Cd
DASH wallet - XtzYFYDPCNfGzJ1z3kG3eudCwdP9fj3fyE

📚📚 My free online machine learning lectures

🧾🧾More insights on NGC

A selection of GPU-optimized applications for AI, HPC, and visualization may be found in the NGC Catalog. It allows you to manage containers. Containers are self-contained environments that bundle software programs, libraries, dependencies, and run-time compilers to make them simple to deploy across diverse computing environments. Through a single command, users may retrieve, operate, and scale programs across the cloud, the datacenter, and the edge, enabling software portability. Data scientists and developers may utilize the pre-trained models for inference or fine-tune them with transfer learning, saving them important time. With the most up-to-date accuracy to enable repeatability, resources include reference neural network topologies across all domains and well-liked frameworks, as well as documentation and code examples that make it simple to get started with deep learning. If you’re a Kubernetes user, well NGC is perfect for you. WHY ? The deployment and maintenance of containerized apps and microservices are made easier by the container orchestrator Kubernetes. DevOps may more quickly setup, deploy, and update apps across Kubernetes environments using a package management called a Helm chart. For the deployment of apps and SDKs that are GPU-optimized, the NGC Catalog offers Helm charts. Are you into SDKs ? All the tools required to create and deploy AI applications across domains like medical imaging, conversational AI, or video analytics are provided through SDKs.

🧾🧾More insights on Vertex AI Workbench Google Cloud

#AI #computervision #JupyterNotebooks

[fatihtarikata]

Very underrated video, Ahmad is pushing Google products, I mean this guys should be recommended everywhere.

[ahkajans]

I am in MLOPs and Google’s Vertex AI is one platform-as-a-service for machine learning and AI developers to encourage MLOps in the industry that is super highly recommended.

[ahmedalahly.]

I used to use openCV to extract segments from vdo, but this video is on another level.

[minecrftlegend]

Highly recommended video! A hidden gem! Suggest beginners to pay attention it will save time.👍👍👍

[ibrahimglr.]

Sir you are awesome. i regret not visiting this channel earlier. Thanks a lot sir.

[محمد_العراقي-بر]

you and your videos deserve more attention than that. you are super hero of engineering student. thank

[twitchssbaba]

Just getting started with Google Cloud. This video was a great help. Thanks so much.

[yonxfb]

Yes. Custom object detection with any of the 40 models tutorial is planned for next week.

[cagdasulucanl]

Thanks again for taking the time to post such detailed videos..

[clgnoyunyt]

tremendous upload Ahmad. I broke that thumbs up on your video. Keep up the brilliant work.

[username]

Thank you for sharing your work with us Ahmad !

[ramin]

Hi there, finally the best tutorial combining Python's OOP and Tensorflow w/ Object detection. Big Thank for that! Very happy to see VS code instead of Jupyter ...

[fghnjgthnyhjh]

Sr you are the best!!!! your video is just like i need. In a close future i will signed in your courses

[joikaygamers]

Thank you very much for this wonderful tutorial it was great help .👏👏😊❤

[iiwiqhwjjaia]

Thanks for the tutorial is very clary! From Patagonia 👏

[cnaryuksel]

Your great content needs to be showed for bigger audience! This of course takes time especially given the biased youtube algorithm.

[klklklkklklkl]

deliated convolution tricks used in deeplab models can detect certain features at different scales.

[emrebeys]

Start watching you and you are awesome ✨ ❤

[lilhase]

Great video!! Why don't you use pyinstaller for exe packaging?

[wraqrwqrwq]

Excellent tutorial. Very neat and straight up to the point.