Deep Learning (Wk 3)

Pre-Trained Models (Example from Kaggle and Book)

Agenda

Play the Piano (link)

Data Augmentation

MNIST (.csv file) — Kaggle Tutorial (link) + My Notebook (link)

Work on Pre-Trained Models (Tutorials) (Hot Dog or Not)

Work on Hand Symbols

Slideshare View

• Cycle GAN: Applied to neural transfer style. For example, you can turn a horse into a zebra or a Monet painting into one that appears to come from van Gough. By exploring the project at https://github.com/junyanz/CycleGAN, you can see how it works and consider the kind of transformations it can apply to images.
• Super Resolution GAN (SRGAN): Transforms images by making blurred, low-resolution images into clear, high-resolution ones. The application of this technique to photography and cinema is interesting because it improves low-quality images at nearly no cost. You can find the paper describing the technique and the results here: https://arxiv.org/pdf/1609.04802.pdf.
• Pose Guided Person Image Generation: Controls the pose of the person depicted in the created image. The paper at https://arxiv.org/pdf/1705.09368.pdf describes practical uses in the fashion industry to generate more poses of a model, but you might be surprised to know that the same approach can create videos of one person dancing exactly the same as another one: https://www.youtube.com/watch?v=PCBTZh41Ris
• Pix2Pix: Translates sketches and maps into real images and vice versa. You can use this application to transform architectural sketches into a picture of a real building or to convert a satellite photo into a drawn map. The paper at https://arxiv.org/pdf/1611.07004.pdf discusses more of the possibilities offered the Pix2Pix network.
• Image repairing: Repairs or modifies an existing image by determining what’s missing, cancelled, or obscured: https://github.com/pathak22/context-encoder.
• Face Aging: Determines how a face will age. You can read about it at https://arxiv.org/pdf/1702.01983.pdf.
• Midi Net: Creates music in your favorite style, as described at https://arxiv.org/pdf/1703.10847.pdf.

[Source: Deep Learning for Dummies]

https://teachablemachine.withgoogle.com/train/image/1q_2sMKdPsrVyfCvT-ASwnkLOmb60pnIV