TensorFlow Lite for Edge Devices - Tutorial

Learn how to use TensorFlow Lite. TensorFlow Lite is an open source deep learning framework for on-device inference.

⭐️ Course Contents ⭐️
(0:00:00) Introduction
(0:00:20) Why do we need TensorFlow Lite?
(0:04:10) What is Edge Computing?
(0:05:06) Why is Edge Computing gaining popularity?
(0:08:30) Challenges in deploying models on Edge devices
(0:09:27) What is TensorFlow Lite or TFLite?
(0:10:06) TensorFlow Lite Workflow
(0:13:30) Creating a TensorFlow or Keras model
(0:32:54) Converting a TensorFlow or Keras model to TFLite
(0:36:08) Validating the TFLite model performance
(0:46:08) What is Quantization?
(0:47:35) Compressing the TFLite model further
(0:51:55) Compressing the TFLite model even further
(0:55:08) Validating the most compressed TFLite model performance
(0:58:46) Thank You

🎉 Thanks to our Champion and Sponsor supporters:
👾 Wong Voon jinq
👾 hexploitation
👾 Katia Moran
👾 BlckPhantom
👾 Nick Raker
👾 Otis Morgan
👾 DeezMaster
👾 AppWrite

--

[shivensaini]

Man I feel so blessed, That I am born in such a good time of humanity.

[ktrivikram]

We wanted to do a project in college in that includes tensorflow and arduino or rasberry pi but i was worried that they could not handle tensorflow so we decided to use it as slave device while a laptop does all the processing, but now i am more than happy seeing this.

[ldanielfch]

Thank you so much! I am doing a researching in my degree about this topic and I just watched this entire video

[adityakharbanda]

Lovely tutorial! Would love to see a tutorial of building an app on iOS to deploy any TFLite models. Thanks!

[Alcatraz-uiyq]

Thank you for the amazing course, it was well structed and well explained! Amazing what quantization can achieve! 😎

[Harshith.N-ep]

Amazing content, gives a complete overview on tflite..!!

[programmingstudy]

nice teachers and nice channel for both creaters and students....

[adayinthelife]

Great tutorial. I like how you show all your dependencies right away, but you don't mention anything about the edge device. I think all the libraries we need to install to run this model will make model compression seem pointless.

[madhumithaa]

I am wondering, How can someone be so perfect and clear in explaining. Crystal Clear Explanation! This really helped me a lot. I request you to do lot many more videos like this.

[hichamm]

It will be better to show how to use the tflite model with a camera by a bib like opencv

[sharanh]

How do we do it for int8 object detection? can you explain this also.

[roku-device]

An Arduino Micro-controller is not an edge device nor is a cell phone. A cell phone can be an edge client, and a micro-controller needs to be connected to a SOC or computer to be an edge device. I am surprised you didn't list a raspberry pi as an edge device because most edge devices are raspberry pi.

[wuzhao]

What is the accuracy for the float16 case?

[OlehBedrii]

Aaaand it's time for me to dust off my old RPi3!

[interspecies-au]

How can run this tflite model in esp8266?

[sanjaybhatikar]

What happens if I constrain the weights to be integers during the training process itself? Does the optimization algorithm still converge as when the weights were floats?

[ocatal]

Thanks for the video!
Can I use TF Lite for a dataset with 12k samples to deploy to mobile? Is there any limitations or does this situation affect the metrics such as acc or something else?

[mcfrenzyo]

Hi Bhavesh Bhatt. Please have you got any idea why my confusion matrix and classification report are only classifying the "0" label and showing 0 classification for "1" label? The dataset is not imbalance as the 0 label are about 120, 000 and the 1 label are about 97, 000. I have also used your code verbatim and still got only the 0 label classified.

[sarath]

Can anyone tell me where to use a Tensor Processing Unit (TPU)?

[Kjy]

Hey free code camp team, Please try to upload a rust programming course, for building decentralized applications and Smart Contract on Solana just like you uploaded a solidity course for Ethereum. Thanks!