models/autoencoder.py

import torch
import torch.nn as nn

# ---------------------------------------------------------------------------
# Original Denoising Autoencoder (96×96 → 96×96)
# ---------------------------------------------------------------------------

class Encoder(nn.Module):
    """Convolutional encoder: 3×96×96 → 256×12×12"""

    def __init__(self):
        super().__init__()
        self.net = nn.Sequential(
            # 3×96×96 → 32×96×96
            nn.Conv2d(3, 32, kernel_size=3, padding=1),
            nn.BatchNorm2d(32),
            nn.ReLU(inplace=True),

            # 32×96×96 → 64×48×48
            nn.Conv2d(32, 64, kernel_size=3, stride=2, padding=1),
            nn.BatchNorm2d(64),
            nn.ReLU(inplace=True),

            # 64×48×48 → 128×24×24
            nn.Conv2d(64, 128, kernel_size=3, stride=2, padding=1),
            nn.BatchNorm2d(128),
            nn.ReLU(inplace=True),

            # 128×24×24 → 256×12×12
            nn.Conv2d(128, 256, kernel_size=3, stride=2, padding=1),
            nn.BatchNorm2d(256),
            nn.ReLU(inplace=True),
        )

    def forward(self, x):
        return self.net(x)


class Decoder(nn.Module):
    """Convolutional decoder: 256×12×12 → 3×96×96"""

    def __init__(self):
        super().__init__()
        self.net = nn.Sequential(
            # 256×12×12 → 128×24×24
            nn.ConvTranspose2d(256, 128, kernel_size=3, stride=2, padding=1, output_padding=1),
            nn.BatchNorm2d(128),
            nn.ReLU(inplace=True),

            # 128×24×24 → 64×48×48
            nn.ConvTranspose2d(128, 64, kernel_size=3, stride=2, padding=1, output_padding=1),
            nn.BatchNorm2d(64),
            nn.ReLU(inplace=True),

            # 64×48×48 → 32×96×96
            nn.ConvTranspose2d(64, 32, kernel_size=3, stride=2, padding=1, output_padding=1),
            nn.BatchNorm2d(32),
            nn.ReLU(inplace=True),

            # 32×96×96 → 3×96×96
            nn.Conv2d(32, 3, kernel_size=3, padding=1),
            nn.Sigmoid(),
        )

    def forward(self, x):
        return self.net(x)


class DenoisingAutoencoder(nn.Module):
    """Full denoising autoencoder: noisy image → clean image."""

    def __init__(self):
        super().__init__()
        self.encoder = Encoder()
        self.decoder = Decoder()

    def forward(self, x):
        z = self.encoder(x)
        return self.decoder(z)

    def count_parameters(self):
        return sum(p.numel() for p in self.parameters() if p.requires_grad)


# ---------------------------------------------------------------------------
# Super-Resolution Autoencoder (noisy 48×48 → clean 96×96)
# ---------------------------------------------------------------------------

class SREncoder(nn.Module):
    """Convolutional encoder: 3×48×48 → 256×6×6"""

    def __init__(self):
        super().__init__()
        self.net = nn.Sequential(
            # 3×48×48 → 32×48×48
            nn.Conv2d(3, 32, kernel_size=3, padding=1),
            nn.BatchNorm2d(32),
            nn.ReLU(inplace=True),

            # 32×48×48 → 64×24×24
            nn.Conv2d(32, 64, kernel_size=3, stride=2, padding=1),
            nn.BatchNorm2d(64),
            nn.ReLU(inplace=True),

            # 64×24×24 → 128×12×12
            nn.Conv2d(64, 128, kernel_size=3, stride=2, padding=1),
            nn.BatchNorm2d(128),
            nn.ReLU(inplace=True),

            # 128×12×12 → 256×6×6
            nn.Conv2d(128, 256, kernel_size=3, stride=2, padding=1),
            nn.BatchNorm2d(256),
            nn.ReLU(inplace=True),
        )

    def forward(self, x):
        return self.net(x)


class SRDecoder(nn.Module):
    """Convolutional decoder: 256×6×6 → 3×96×96 (2× upscale)"""

    def __init__(self):
        super().__init__()
        self.net = nn.Sequential(
            # 256×6×6 → 128×12×12
            nn.ConvTranspose2d(256, 128, kernel_size=3, stride=2, padding=1, output_padding=1),
            nn.BatchNorm2d(128),
            nn.ReLU(inplace=True),

            # 128×12×12 → 64×24×24
            nn.ConvTranspose2d(128, 64, kernel_size=3, stride=2, padding=1, output_padding=1),
            nn.BatchNorm2d(64),
            nn.ReLU(inplace=True),

            # 64×24×24 → 32×48×48
            nn.ConvTranspose2d(64, 32, kernel_size=3, stride=2, padding=1, output_padding=1),
            nn.BatchNorm2d(32),
            nn.ReLU(inplace=True),

            # 32×48×48 → 16×96×96  ← extra layer gives the 2× upscale
            nn.ConvTranspose2d(32, 16, kernel_size=3, stride=2, padding=1, output_padding=1),
            nn.BatchNorm2d(16),
            nn.ReLU(inplace=True),

            # 16×96×96 → 3×96×96
            nn.Conv2d(16, 3, kernel_size=3, padding=1),
            nn.Sigmoid(),
        )

    def forward(self, x):
        return self.net(x)


class SuperResAutoencoder(nn.Module):
    """Denoise + 2× upscale: noisy 48×48 → clean 96×96.

    Fully convolutional — can accept any input size and will output 2× that size.
    """

    def __init__(self):
        super().__init__()
        self.encoder = SREncoder()
        self.decoder = SRDecoder()

    def forward(self, x):
        return self.decoder(self.encoder(x))

    def count_parameters(self):
        return sum(p.numel() for p in self.parameters() if p.requires_grad)