Jovian
Sign In
Learn practical skills, build real-world projects, and advance your career

#Basics and Ideas for improvement

###Tips on Training GANs:

Upsampling vs Transposed Convolution:

  • Upsampling looking promising, less chessboard artefacts.

Input Noise:

  • Noisevector at 128 or 256 size

Batchsize:

  • small Batchsizes faster and often better for the Discriminator b < 64 (8 or 16 to try)

Loss Function:

  • trying different Lossfunctions such as Wasserstein or RaLS

##To Do List:

  • fix save model

  • get a way to load model

  • different hyperparameter (lr, leaky Relu)

  • start with only black or white noise

  • different Loss Function

  • bigger images size (more / less CNN layer)

  • change Inputnoisevector

  • try batchsize of 8 or 16

  • identify and change Convolution method

source: https://medium.com/intel-student-ambassadors/tips-on-training-your-gans-faster-and-achieve-better-results-9200354acaa5

#Basic Setup

import os
import torch
import torchvision
import tarfile
from torchvision.datasets.utils import download_url
from torch.utils.data import random_split
project_name = "zerotogans-project-aaron-wagner"
path = "/content/input"

try:
    os.mkdir(path)
except OSError:
    print ("Creation of the directory %s failed" % path)
else:
    print ("Successfully created the directory %s " % path)
Successfully created the directory /content/input