125 - What are Generative Adversarial Networks (GAN)?

Generative adversarial networks (GANs) are deep learning architectures that use two neural networks (Generator and Discriminator), competing one against the other. The generator tries to create realistic looking fake data (e.g. images) and the discriminator tries to classify whether the data is real or fake. After a few thousand (or million) epochs, the generator trained model can be used to create new fake data that can pass for real data.

This tutorial provides a quick overview of GANs. The next tutorial in the playlist covers the implemetation of GAN using Keras in Python.

References from the video:

[sriharimohan]

one of the best channels for Deep Learning in Images. Thank you Sir for these wonderful tutorials

[snekisnowhite]

I watched at least 10 videos on GAN, this one cleared my mind the best what is happening in GAN how it actually works..

[me-ourf]

Can GAN be good for Data Augmentation for EEG ?

[ExV]

Thank you so much sir. You teach much better than my professor.

[peshawriankhan]

Hi Dear sir
Is there is Any practicle project on GAN, s in your video list?with coding?

[mahmoodkashmiri]

After watching your videos I feel confident enough to create something amazing! thank you

[sondosmahd]

can you make a video with using GAN to detect text not image (let say as ex: attack text & not attack text for site), where discriminator contain 2 layer?

[sgrimm]

So, it sounds to me like random noise is the input to the generator while the discriminator contains the 'target' information, let's say an image. The generator network is trained using the discriminator data until the error loss is acceptable. Correct? If this is true, how is this any different from a standard ANN that is trained via supervision? Or am I getting something wrong? I'm trying to figure this out. Thank you. Good info.

[BareqRaad]

Thanks for this great demonstration
Sir am trying to locate the forged part of an image which deep learning architecture you advice me to work on

[Induraj]

Nice and crystal clear explanation. keep continuing sir

[SwarnaliMollickA]

Sir, my GPU is NVIDIA GeForce RTX 2060 and I have 32GB RAM. Is it enough to work with GANs? Please reply Sir.

[Srimathyamutha]

Will GAN be helpful in repairing broken letters in images after pre processing them for OCR ?

[johnpuskin]

Hi. Can you compare classical upsampling based high resolution image generation DNNs with SR-GAN? When and why we should prefer GANs?

[salmankhalildurrani]

Great video, very practical. Keep sending more!

[gamerpunk]

A great video sir! Thank you soo much for the crystal clear explanation!

[samanehimani]

thanks very much for good teaching. i also watched variational autoencoder and it was perfect.

[penugondasaichand]

can we say if loss is getting low then fake images is not generated and if loss is getting higher then fake images are generated ???when we have given noise data and image file to gans????/

[microcosmos]

As always, very clear explanation, thanks!

[mimo-wxmc]

thank u so much sir, can you do a video on denoising ct images using generative adversarial networks

[lequedicatsamarge]

Would it make sense to use an VAE as a generator and then train the discriminator based in the input- vs output data of the VAE? And I wonder if I could use the trained discriminator for anomaly detection.
The thing is, I have acoustic data of a running machine that has never failed (and it should not, it is a giant 100 kil-tons steel wheel rotating at high-speed) and I want to model an early-warning-system. It seems like the discriminator would be a tool that can be used for this, since the data overall is fairly homogeneous.