Can't get VAE loss (on MNIST) below 0.92

Mehdi · April 1, 2018, 2:15pm

Hello,

I wrote a vae (heavily inspired from the Examples) to try and generate MNIST digits.
However, I can’t get the global loss below 0.92.

Here’s a link for the code:

github.com

Mehd6384/Generative-Networks/blob/master/vae.py

import torch 
import torch.nn as nn 
import torch.nn.functional as F 
import torch.optim as optim	 
from torch.autograd import Variable
from torch.distributions import Normal 

from torch.utils.data import DataLoader
from torchvision.datasets import MNIST as mnist
from torchvision import transforms

import numpy as np 
import matplotlib.pyplot as plt 
plt.style.use('ggplot')

batch_size = 32 

transform = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.1307,), (0.3081,))])
data = DataLoader(mnist('../data', train = True, download = False, transform = transform), shuffle = True, batch_size = batch_size)

This file has been truncated. show original

Is there something I missed ?

Thanks !