import streamlit as st import torch import torch.nn as nn from torchvision import transforms from PIL import Image import numpy as np class ResidualBlock(nn.Module): def __init__(self, channels): super().__init__() self.block = nn.Sequential( nn.ReflectionPad2d(1), nn.Conv2d(channels, channels, 3), nn.InstanceNorm2d(channels), nn.ReLU(inplace=True), nn.ReflectionPad2d(1), nn.Conv2d(channels, channels, 3), nn.InstanceNorm2d(channels), ) def forward(self, x): return x + self.block(x) class ResNetGenerator(nn.Module): def __init__(self, in_ch=3, out_ch=3, ngf=32, n_blocks=6): super().__init__() layers = [ nn.ReflectionPad2d(3), nn.Conv2d(in_ch, ngf, 7), nn.InstanceNorm2d(ngf), nn.ReLU(inplace=True), ] mult = 1 for i in range(2): layers += [ nn.Conv2d(ngf * mult, ngf * mult * 2, 3, stride=2, padding=1), nn.InstanceNorm2d(ngf * mult * 2), nn.ReLU(inplace=True), ] mult *= 2 for _ in range(n_blocks): layers.append(ResidualBlock(ngf * mult)) for i in range(2): layers += [ nn.ConvTranspose2d(ngf * mult, ngf * mult // 2, 3, stride=2, padding=1, output_padding=1), nn.InstanceNorm2d(ngf * mult // 2), nn.ReLU(inplace=True), ] mult //= 2 layers += [nn.ReflectionPad2d(3), nn.Conv2d(ngf, out_ch, 7), nn.Tanh()] self.model = nn.Sequential(*layers) def forward(self, x): return self.model(x) @st.cache_resource def load_models(): device = torch.device("cpu") checkpoint = torch.load("cyclegan_dogs2cats_final.pt", map_location=device) prefix_ab = "generators.a_to_b." prefix_ba = "generators.b_to_a." state_ab = {k[len(prefix_ab):]: v for k, v in checkpoint.items() if k.startswith(prefix_ab)} state_ba = {k[len(prefix_ba):]: v for k, v in checkpoint.items() if k.startswith(prefix_ba)} gen_ab = ResNetGenerator(ngf=32, n_blocks=6) gen_ba = ResNetGenerator(ngf=32, n_blocks=6) gen_ab.load_state_dict(state_ab) gen_ba.load_state_dict(state_ba) gen_ab.eval() gen_ba.eval() return gen_ab, gen_ba def preprocess(image): transform = transforms.Compose([ transforms.Resize((128, 128)), transforms.ToTensor(), transforms.Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5]), ]) return transform(image).unsqueeze(0) def postprocess(tensor): img = tensor.squeeze(0).detach().cpu().numpy() img = (img * 0.5 + 0.5).clip(0, 1) img = (img.transpose(1, 2, 0) * 255).astype(np.uint8) return Image.fromarray(img) def main(): st.set_page_config(page_title="CycleGAN: Dogs & Cats", layout="centered") st.title("CycleGAN: Dogs ↔ Cats") gen_ab, gen_ba = load_models() direction = st.radio("Направление трансформации", ["Dog → Cat", "Cat → Dog"]) uploaded = st.file_uploader("Загрузите изображение", type=["jpg", "jpeg", "png", "webp", "bmp"]) if uploaded is not None: image = Image.open(uploaded).convert("RGB") col1, col2 = st.columns(2) col1.subheader("Оригинал") col1.image(image, use_container_width=True) inp = preprocess(image) with torch.no_grad(): if direction == "Dog → Cat": out = gen_ab(inp) else: out = gen_ba(inp) result = postprocess(out) col2.subheader("Результат") col2.image(result, use_container_width=True) if __name__ == "__main__": main()