Hi,

Back in high school, one of my friends dared me to chug a bottle of hot sauce during lunch.

He offered me $5.

And as a broke high schooler in need of gas money, it seemed like a good idea.

So I took the $5, grabbed the bottle of hot sauce, and chugged it down like a champ.

Except ... back in high school I weighed a whopping 95lbs ... so you can imagine what that bottle of hot sauce did to my intestines.

I spent the second half of the school day in the bathroom, praying to the porcelain gods to rescue me and bring me salvation.

Image


Salvation never came, I missed the bus home, and I had to walk 1.5 miles back to my house through cornfields, occasionally stopping to ralf up some foul red looking mixture of hot sauce and bile.

Fun times.

Why am I telling you this story? And what does it have to do with tf.data and deep learning?

I'm getting there ...

The big picture: I never thought I would find something that could go faster than hot sauce through my lower intestine — but my search is over. I've found it.

The tf.data data pipeline module is so fast that you'll think your neural network just chugged a bottle of ghost pepper hot sauce and then made a beeline for the bathroom.

In today's brand new tutorial I show you how to use the tf.data module to train a CNN on image data in record time:

Image

How it works: The tf.data pipeline is > 4x faster than using Keras' standard ImageDataGenerator.

Under the hood, tf.data optimizes the entire data processing pipeline, giving you blazing fast processing speeds with little overhead.

Yes, but: As I mentioned last week, one of the downsides to tf.data is that it can make data augmentation a bit more challenging to apply. Next week I'll discuss how to easily incorporate data augmentation into your blazing fast tf.data pipeline.

Stay smart: Don't chug hot sauce straight from the bottle. Do use tf.data when training your own CNNs.

Click here to learn how to train your own CNNs using the super fast tf.data pipelines.
 
Adrian Rosebrock
Chief PyImageSearcher

P.S. Have a question about Computer Vision, Deep Learning, and OpenCV?

I'd be happy to help — but the fact of the matter is that I receive 200+ emails per day asking for a bit of my time. There's simply not enough time in the day for me to respond to these emails and still have a happy, productive, and sustainable life.

Over the past 5 years I've authored over 350 free tutorials to learn from — and while it's been an absolute pleasure — if you want a bit of my time, I have to politely ask that you purchase one of my books or courses.

Not only will you be getting a great education through my books/courses, but if you have a question after becoming a customer, I will ensure it is answered.

If you have any questions on which book/course is right for you, just reply to this email, and I'll put you in touch with my Enrollment Advisor. They will personally guide you to which book/course is best for you and your project.