Generative-Adversarial Networks [COMPUTER-VISION]

For documentation please run helpme.py

This model generate images and Discrement them on the basis of real images, gives surprising results, can apply arithmatic operations on generated images.

NOTE : You can tweak the parameter of training dataset, kernel, and filter functions (RELU can be replaced with LSTM for text generation Process)

Installation

1. install pillow, numpy and mpi4py
2. install cuda (10.2) version of pytrorch and torchvision

PreRequisite

1. Knowlege of Python (Basic[class, modules, function])
2. Pytorch, Neural Network
3. Activation Function.

To Train and Get Generated Images Run design.py

Cool you are good to go here is some Definition to help you:

PAPER :
  GAN, MD-GAN :
    Generative Adversarial Network, Multi-Discriminator Generative Adversarial Networks for Distributed Datasets

	
AUTHORS	:

  PAPER I - [ Ian J. Goodfellow,  Jean Pouget-Abadie∗, Mehdi Mirza, Bing Xu, David Warde-Farley,Sherjil Ozair†, Aaron      Courville, Yoshua Bengio‡ ]
  PAPER II - [ Corentin Hardy, Erwan Le Merrer, Bruno Sericola ]

  LINK :
            PAPER I :  https://arxiv.org/pdf/1406.2661.pdf
            PAPER II : https://arxiv.org/pdf/1811.03850v2.pdf

************************************************************************************************************************************************************************************


Main libraries :

	numpy			:	It's a multidimensional Array.

	mpi4py			:	MPI for Python supports convenient, pickle-based communication of generic Python object as well as fast, near C-speed, direct array data communication of buffer-provider objects.

	torch			:	An open source machine learning framework that accelerates the path from research prototyping to production deployment.(Official site)	

	torch.nn:		:	Base class for all neural network modules, our models is also subclass this class.		

	torch.optim		:	torch.optim is a package implementing various optimization algorithms. Most commonly used methods are already supported, and the interface is general enough,
					 so that more sophisticated ones can be also easily integrated in the future.

	torch.utils.data	:	It represents a python iterable over a dataset.

	torch.nn.parallel	:	This container parallelizes the application of the given module by splitting the input across the specified devices by chunking in the batch dimension.

	torchvision		:	The torchvision package consists of popular datasets, model architectures, and common image transformations for computer vision.

Other libraries :

	random, os.


************************************************************************************************************************************************************************************


Important Variables :

	size			:	Represt the size of message pass to suffle the Discriminator using peer2peer fashion.

	rank			:	Push the Discriminator to use respective position (bcz we are using two discriminator, rank discriminator by 1/0 (1 : run next, 0: currently running))
	
	datasets		:	This varible is heart of the programme bcz this will download and store the dataset. 

	dataloader		:	At the heart of PyTorch data loading utility is the torch.utils.data.DataLoader class, We are using one HUGE DATASET CIFAR10 in this implimentation.



************************************************************************************************************************************************************************************


Main Class :

	G()			:	This class is used to create generator Neural Network by using torch.nn.Module class.

	D()			:	This class is used to create Discriminator Neural Network by using torch.nn.Module class.


************************************************************************************************************************************************************************************



Main function :


	copyGenerator		:	Create a copy of generator to get the feedback of the generator to learn from them.

	shuffleDiscriminators	:	Shuffle the discriminator on the basis of rank after every 2 epochs.

THESE ARE SPECIAL FUNCTION USED FOR MD-GAN


************************************************************************************************************************************************************************************
Sudo code of implimentation of MD-GAN :

	Algorithm  1MD-GAN algorithm
			1:procedureWORKER(C,Bn,I,L,b)
			2:	InitializeθnforDn
			3:	fori←1toIdo
			4:		X(r)n←SAMPLES(Bn,b)
			5:		X(g)n,X(d)n←RECEIVEBATCHES(C)
			6:		forl←0toLdo
			7:			Dn←DISCLEARNINGSTEP(Jdisc,Dn)
			8:		end for
			9:	Fn←{∂ ̃B(X(g)n)∂xi|xi∈X(g)n}
			10:	SEND(C,Fn).SendFnto server
			11:	ifimod (mEb) = 0then
			12:		Dn←SWAP(Dn)
			13:	end if
			14:	end for
			15:end procedure
			16:
			17:procedureSWAP(Dn)
			18:	Wl←GETRANDOMWORKER()
			19:	SEND(Wl,Dn).SendDnto workerWl.
			20:	Dn←RECEIVED().Receive a new discriminatorfrom another worker.
			21:	ReturnDn
			22:end procedure
			23:
			24:procedureSERVER(k,I).Server C
			25:	InitializewforG
			26:	fori←1toIdo
			27:		forj←0tokdo
			28:			Zj←GAUSSIANNOISE(b)
			29:			X(j)←{Gw(z)|z∈Zj}
			30:		end for
			31:		X(d)1,...,X(d)n←SPLIT(X(1),...,X(k))
			32:		X(g)1,...,X(g)n←SPLIT(X(1),...,X(k))
			33:		for n←1toNdo
			34:			SEND(Wn,(X(d)n,X(g)n))
			35:		end for
			36:		F1,...,FN←GETFEEDBACKFROMWORKERS()
			37:		Compute∆waccording toF1,...,FN
			38:		for wi∈w do
			39:			wi←wi+ADAM(∆wi)
			40:		end for
			41:	end for
			42:end procedure

---

The results are really excited :

These are the generated image -

Epoch 0

• 

Epoch 1

• 

Epoch 10

• 

Epoch 20

• 

Epoch 24

• 

Real Image

• 

Real Image(LEFT) VS Generated Image(RIGHT)

.....