nota_wav2lip/models/wav2lip.py

import torch
from torch import nn

from nota_wav2lip.models.base import Wav2LipBase
from nota_wav2lip.models.conv import Conv2d, Conv2dTranspose


class Wav2Lip(Wav2LipBase):
    def __init__(self):
        super().__init__()

        self.face_encoder_blocks = nn.ModuleList([
            nn.Sequential(Conv2d(6, 16, kernel_size=7, stride=1, padding=3)), # 96,96

            nn.Sequential(Conv2d(16, 32, kernel_size=3, stride=2, padding=1), # 48,48
            Conv2d(32, 32, kernel_size=3, stride=1, padding=1, residual=True),
            Conv2d(32, 32, kernel_size=3, stride=1, padding=1, residual=True)),

            nn.Sequential(Conv2d(32, 64, kernel_size=3, stride=2, padding=1),    # 24,24
            Conv2d(64, 64, kernel_size=3, stride=1, padding=1, residual=True),
            Conv2d(64, 64, kernel_size=3, stride=1, padding=1, residual=True),
            Conv2d(64, 64, kernel_size=3, stride=1, padding=1, residual=True)),

            nn.Sequential(Conv2d(64, 128, kernel_size=3, stride=2, padding=1),   # 12,12
            Conv2d(128, 128, kernel_size=3, stride=1, padding=1, residual=True),
            Conv2d(128, 128, kernel_size=3, stride=1, padding=1, residual=True)),

            nn.Sequential(Conv2d(128, 256, kernel_size=3, stride=2, padding=1),       # 6,6
            Conv2d(256, 256, kernel_size=3, stride=1, padding=1, residual=True),
            Conv2d(256, 256, kernel_size=3, stride=1, padding=1, residual=True)),

            nn.Sequential(Conv2d(256, 512, kernel_size=3, stride=2, padding=1),     # 3,3
            Conv2d(512, 512, kernel_size=3, stride=1, padding=1, residual=True),),

            nn.Sequential(Conv2d(512, 512, kernel_size=3, stride=1, padding=0),     # 1, 1
            Conv2d(512, 512, kernel_size=1, stride=1, padding=0)),])

        self.audio_encoder = nn.Sequential(
            Conv2d(1, 32, kernel_size=3, stride=1, padding=1),
            Conv2d(32, 32, kernel_size=3, stride=1, padding=1, residual=True),
            Conv2d(32, 32, kernel_size=3, stride=1, padding=1, residual=True),

            Conv2d(32, 64, kernel_size=3, stride=(3, 1), padding=1),
            Conv2d(64, 64, kernel_size=3, stride=1, padding=1, residual=True),
            Conv2d(64, 64, kernel_size=3, stride=1, padding=1, residual=True),

            Conv2d(64, 128, kernel_size=3, stride=3, padding=1),
            Conv2d(128, 128, kernel_size=3, stride=1, padding=1, residual=True),
            Conv2d(128, 128, kernel_size=3, stride=1, padding=1, residual=True),

            Conv2d(128, 256, kernel_size=3, stride=(3, 2), padding=1),
            Conv2d(256, 256, kernel_size=3, stride=1, padding=1, residual=True),

            Conv2d(256, 512, kernel_size=3, stride=1, padding=0),
            Conv2d(512, 512, kernel_size=1, stride=1, padding=0),)

        self.face_decoder_blocks = nn.ModuleList([
            nn.Sequential(Conv2d(512, 512, kernel_size=1, stride=1, padding=0),),

            nn.Sequential(Conv2dTranspose(1024, 512, kernel_size=3, stride=1, padding=0), # 3,3
            Conv2d(512, 512, kernel_size=3, stride=1, padding=1, residual=True),),

            nn.Sequential(Conv2dTranspose(1024, 512, kernel_size=3, stride=2, padding=1, output_padding=1),
            Conv2d(512, 512, kernel_size=3, stride=1, padding=1, residual=True),
            Conv2d(512, 512, kernel_size=3, stride=1, padding=1, residual=True),), # 6, 6

            nn.Sequential(Conv2dTranspose(768, 384, kernel_size=3, stride=2, padding=1, output_padding=1),
            Conv2d(384, 384, kernel_size=3, stride=1, padding=1, residual=True),
            Conv2d(384, 384, kernel_size=3, stride=1, padding=1, residual=True),), # 12, 12

            nn.Sequential(Conv2dTranspose(512, 256, kernel_size=3, stride=2, padding=1, output_padding=1),
            Conv2d(256, 256, kernel_size=3, stride=1, padding=1, residual=True),
            Conv2d(256, 256, kernel_size=3, stride=1, padding=1, residual=True),), # 24, 24

            nn.Sequential(Conv2dTranspose(320, 128, kernel_size=3, stride=2, padding=1, output_padding=1),
            Conv2d(128, 128, kernel_size=3, stride=1, padding=1, residual=True),
            Conv2d(128, 128, kernel_size=3, stride=1, padding=1, residual=True),), # 48, 48

            nn.Sequential(Conv2dTranspose(160, 64, kernel_size=3, stride=2, padding=1, output_padding=1),
            Conv2d(64, 64, kernel_size=3, stride=1, padding=1, residual=True),
            Conv2d(64, 64, kernel_size=3, stride=1, padding=1, residual=True),),]) # 96,96

        self.output_block = nn.Sequential(Conv2d(80, 32, kernel_size=3, stride=1, padding=1),
            nn.Conv2d(32, 3, kernel_size=1, stride=1, padding=0),
            nn.Sigmoid())