TensorFlow Tutorial 11 - Transfer Learning, Fine Tuning and TensorFlow Hub

Download TensorFlow Hub (w. pip it's pip install tensorflow-hub):

❤️ Support the channel ❤️

Paid Courses I recommend for learning (affiliate links, no extra cost for you):

✨ Free Resources that are great:

💻 My Deep Learning Setup and Recording Setup:

GitHub Repository:

✅ One-Time Donations:

▶️ You Can Connect with me on:

[AladdinPersson]

I was inspired and learned the basics of TensorFlow after I completed the TensorFlow specialization on coursera. Personally I think these videos I created give a similar understanding but if you wanna check it out you can. Below you'll find both affiliate and non-affiliate links, the pricing for you is the same but a small commission goes back to the channel if you buy it through the affiliate link which helps me create more future videos.

[sulavojha]

Add these on your list : estimator api, TFRrecords, TFX, tensorflow serving, TensorRt

[RajeshPachaikani]

Thank you for this awesome tutorial. Please do segmentation model retraining in the future.

[andreasenswe]

Tusen tack för denna serie. Tack vare dig kom jag igång med tensorflow. ❤️

[saatweek]

Thanks a lot for the tutorial. This was really insightful

[nur__bijoy]

Thanks for this handy tutorial. Well explained brother.

[aliaraslihan]

If you set include_top=True it reinitializes the dense layer with pre-trained weight otherwise it ignores it. If you set include_top=False, the output of the previous layer, passed through a GlobalAvreagePooling2D() that extract most important features.

[maddiesoltani]

Awesome video! But could you please explain what the purpose of like changing the dense layer to 10 nodes instead and what it does or could you please relate it to real life situation like how changing the number of nodes would help?

[hossamalzomor]

thanks for this nice tutorials, I have the following questions
1.did you forget to set the layer of the original model to be non-trainable in keras and tf.hub examples in github code?
2.How many layer should I add after the pre-trained model, and how dense should they be?
3.Is there any difference between loading the model from keras or from tf.hub?
4.Is loading feature_vector equivalent to loading the model and omitting the last layer?

[viiibgw]

I have a question. In this third part(Pretrained Hub Model), Can i frozen some layers in the pre-trained model(just the url downloaded)?

[acebookwisdom]

Thank you for your tutorial.

What would you suggest the right model to use for real time sign language detection?

[kirtisonkar]

Hey, can you please make a video for making feature vectors in pretrained model... Please

[Canada_Travel_life]

hi amazing tutorials. Please include keras tuner installation for pycharm as well.

[Dimitris__]

On the pretrained keras model, can i also change the Input layer?

[haideralishuvo]

Do some Pytorch tutorials on Object detection algorithms like SSD, YOLO, RCNN please

[muhammadzubairbaloch]

such kind of lectures are already availabel on youtube. please it is my request make lectures on read word datasets.

[johnbaker]

Hey! Nice video :D

However,

I get the following error: AttributeError: 'Dense' object has no attribute 'shape'

when trying to call the new output layer as mentioned in your example: any reasons to why?
output =
new_model = keras.Model(base_inputs, output)

[guanquanwang]

Something goes wrong when I use 'import tensorflow_hub as hub'.
The error shows like: "ModuleNotFoundError: No module named 'keras'"
Does anyone meet this problem and fix it successfully?

[mohamedabdul-al]

Hi,
Why when I change the nodes and nb of images and nb of channels it does not work with me!! Are they fixed nbs or what??
I hope you can help me in this.
Thank you so much in advanced.

[a_brodo]

Hi, when I run the pretained keras model with this code, I notice that I have 1/1 below each epoch instead of 5/5? Is this a big issue, because it seems that it's only doing one output layer or something? I believe I've copied your code as it is in the video so is it possible that there's another external issue I have with the software or something? I'm running the latest version of TF from my GPU and have checked that it does detect it properly