Easy to use 100's of Deep Learning models in one place: image, text, video and audio [code included]

TensorFlow Hub is where hundreds of machine learning models come together in one place.

TFHub is the place to easily find the latest ready-to-use models, with documentation, code snippets and more.

1- TFHub's repository of models covers a wide range of machine learning tasks:

• Image: classification, object detection, augmentation and generation
• Text: BERT, ALBERT and many more embeddings to support many natural language understanding tasks such as: Question answering, Classifications and summarization
• Video: Action recognition, and video generation
• Audio: Pitch detection

2- These pre-trained models have been prepared for different environments:

• TensorFlow JS models for web environment
• TensorFlow Lite models for mobile and embedded devices
• Coral models for edge TPU devices

3- Get started with nothing to install:

Many TensorFlow Hub models have interactive Colab notebooks to play with the model with code examples right from the browser

Now let's see some examples using just few lines of code

Neural style transfer (code)

Where we have a content input image and a style input image and want to generate and output image with both the content and style.

BigGAN Demo (code)

BigGAN image generator is also available.

a) Generating images of specific class

b) Interpolation between images

Object detection (code)

Using Faster-RCNN InceptionResNet trained on Open-images dataset.

Text classification (code)

In order to represent the text for neural networks is to convert sentences into embedding vectors, where we can use a pre-trained text embedding as the first layer of our model. Accordingly, we don't have to worry about text preprocessing and benefit from transfer learning.

TFHub as many embeddings for different languages ready for use.

In the code above, the first layer is a TensorFlow Hub layer. This layer uses a pre-trained Saved Model to map a sentence into its embedding vector. 

TFHub is powered by the community: DeepMind, Google, Microsoft AI for Earth, NVIDIA, Kaggle and more ...

References

[1] Blog Introducing TensorFlow Hub

[2] video TensorFlow Hub: Making model discovery easy (TF Dev Summit '20)