import torch import torch.nn as nn import torch.nn.functional as F LATENT_DIM = 128 class VAE(nn.Module): def __init__(self): super().__init__() # Encoder self.enc = nn.Sequential( nn.Conv2d(4, 32, 4, 2, 1), # 64 -> 32 nn.ReLU(), nn.Conv2d(32, 64, 4, 2, 1), # 32 -> 16 nn.ReLU(), nn.Conv2d(64, 128, 4, 2, 1), # 16 -> 8 nn.ReLU(), nn.Flatten() ) self.fc_mu = nn.Linear(128 * 8 * 8, LATENT_DIM) self.fc_logvar = nn.Linear(128 * 8 * 8, LATENT_DIM) # Decoder self.fc_dec = nn.Linear(LATENT_DIM, 128 * 8 * 8) self.dec = nn.Sequential( nn.ConvTranspose2d(128, 64, 4, 2, 1), # 8 -> 16 nn.ReLU(), nn.ConvTranspose2d(64, 32, 4, 2, 1), # 16 -> 32 nn.ReLU(), nn.ConvTranspose2d(32, 4, 4, 2, 1), # 32 -> 64 nn.Sigmoid() ) def encode(self, x): h = self.enc(x) return self.fc_mu(h), self.fc_logvar(h) def reparam(self, mu, logvar): std = torch.exp(0.5 * logvar) eps = torch.randn_like(std) return mu + eps * std def decode(self, z): h = self.fc_dec(z).view(-1, 128, 8, 8) return self.dec(h) def forward(self, x): mu, logvar = self.encode(x) z = self.reparam(mu, logvar) return self.decode(z), mu, logvar