伏羲0.07（文生图）

为了使0.06代码能够有效运行并输出项目目录及所有文件，我们在代码中添加一些额外的功能。

1. 项目目录结构
   项目目录结构如下：

text_to_image_project/
│
├── config.yaml
├── data/
│ ├── train_data.csv
│ └── test_data.txt
├── models/
│ └── text_to_image_model.pth
├── main.py
└── README.md

2. 示例配置文件
   config.yaml

model:path: models/text_to_image_model.pthtext_encoder_model_name: bert-base-uncaseddata:input_file: data/test_data.txtoutput_dir: data/generated_imagesdataset_path: data/train_data.csvtraining:batch_size: 64learning_rate: 0.0002epochs: 100

3. 示例训练数据文件
   data/train_data.csv

text,image_path
a beautiful sunset,data/images/sunset.jpg
a cute puppy,data/images/puppy.jpg
a red rose,data/images/rose.jpg

4. 示例测试数据文件
   data/test_data.txt

a beautiful sunset
a cute puppy
a red rose

5. 完善后的代码
   main.py

import tkinter as tk
from tkinter import filedialog, messagebox
from PIL import Image, ImageTk
import torch
import torch.optim as optim
import torch.nn as nn
import torchvision.transforms as transforms
import yaml
import os
import pandas as pd
from torch.utils.data import Dataset, DataLoader
from transformers import AutoTokenizer, AutoModel
import random
import numpy as np# 配置文件加载
def load_config(config_path):with open(config_path, 'r', encoding='utf-8') as file:config = yaml.safe_load(file)return config# 数据加载
def load_text_data(file_path):with open(file_path, 'r', encoding='utf-8') as file:text_data = file.readlines()return [line.strip() for line in text_data]# 数据清洗
def clean_data(data):return data.dropna().drop_duplicates()# 数据增强
def augment_data(image, mode):if mode == 'train':transform = transforms.Compose([transforms.RandomHorizontalFlip(),transforms.RandomRotation(10),transforms.RandomResizedCrop(64, scale=(0.8, 1.0)),transforms.ColorJitter(brightness=0.2, contrast=0.2, saturation=0.2, hue=0.1),transforms.ToTensor(),transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])else:transform = transforms.Compose([transforms.Resize((64, 64)),transforms.ToTensor(),transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])return transform(image)# 文本编码器
class TextEncoder(nn.Module):def __init__(self, model_name):super(TextEncoder, self).__init__()self.tokenizer = AutoTokenizer.from_pretrained(model_name)self.model = AutoModel.from_pretrained(model_name)def forward(self, text):inputs = self.tokenizer(text, return_tensors='pt', padding=True, truncation=True)outputs = self.model(**inputs)return outputs.last_hidden_state.mean(dim=1)# 图像生成器
class ImageGenerator(nn.Module):def __init__(self, in_channels):super(ImageGenerator, self).__init__()self.decoder = nn.Sequential(nn.ConvTranspose2d(in_channels, 512, kernel_size=4, stride=1, padding=0),nn.BatchNorm2d(512),nn.ReLU(True),nn.ConvTranspose2d(512, 256, kernel_size=4, stride=2, padding=1),nn.BatchNorm2d(256),nn.ReLU(True),nn.ConvTranspose2d(256, 128, kernel_size=4, stride=2, padding=1),nn.BatchNorm2d(128),nn.ReLU(True),nn.ConvTranspose2d(128, 64, kernel_size=4, stride=2, padding=1),nn.BatchNorm2d(64),nn.ReLU(True),nn.ConvTranspose2d(64, 3, kernel_size=4, stride=2, padding=1),nn.Tanh())def forward(self, x):x = x.view(-1, x.size(1), 1, 1)return self.decoder(x)# 判别器
class Discriminator(nn.Module):def __init__(self):super(Discriminator, self).__init__()self.main = nn.Sequential(nn.Conv2d(3, 64, kernel_size=4, stride=2, padding=1),nn.LeakyReLU(0.2, inplace=True),nn.Conv2d(64, 128, kernel_size=4, stride=2, padding=1),nn.BatchNorm2d(128),nn.LeakyReLU(0.2, inplace=True),nn.Conv2d(128, 256, kernel_size=4, stride=2, padding=1),nn.BatchNorm2d(256),nn.LeakyReLU(0.2, inplace=True),nn.Conv2d(256, 512, kernel_size=4, stride=2, padding=1),nn.BatchNorm2d(512),nn.LeakyReLU(0.2, inplace=True),nn.Conv2d(512, 1, kernel_size=4, stride=1, padding=0),nn.Sigmoid())def forward(self, x):return self.main(x)# 模型定义
class TextToImageModel(nn.Module):def __init__(self, text_encoder_model_name):super(TextToImageModel, self).__init__()self.text_encoder = TextEncoder(text_encoder_model_name)self.image_generator = ImageGenerator(768)  # 768 is the hidden size of BERTdef forward(self, text):text_features = self.text_encoder(text)return self.image_generator(text_features)# 模型加载
def load_model(model_path, text_encoder_model_name):model = TextToImageModel(text_encoder_model_name)if os.path.exists(model_path):model.load_state_dict(torch.load(model_path))model.eval()return model# 图像保存
def save_image(image, path):if not os.path.exists(os.path.dirname(path)):os.makedirs(os.path.dirname(path))image.save(path)# 数据集类
class TextToImageDataset(Dataset):def __init__(self, csv_file, transform=None, mode='train'):self.data = pd.read_csv(csv_file)self.data = clean_data(self.data)self.transform = transformself.mode = modedef __len__(self):return len(self.data)def __getitem__(self, idx):text = self.data.iloc[idx]['text']image_path = self.data.iloc[idx]['image_path']image = Image.open(image_path).convert('RGB')if self.transform:image = self.transform(image, self.mode)return text, image# 模型训练
def train_model(config):transform = transforms.Compose([transforms.Resize((64, 64)),transforms.ToTensor(),transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])dataset = TextToImageDataset(config['training']['dataset_path'], transform=augment_data, mode='train')dataloader = DataLoader(dataset, batch_size=config['training']['batch_size'], shuffle=True)model = TextToImageModel(config['model']['text_encoder_model_name'])discriminator = Discriminator()optimizer_g = optim.Adam(model.parameters(), lr=config['training']['learning_rate'])optimizer_d = optim.Adam(discriminator.parameters(), lr=config['training']['learning_rate'])criterion_gan = nn.BCELoss()criterion_l1 = nn.L1Loss()for epoch in range(config['training']['epochs']):model.train()discriminator.train()running_loss_g = 0.0running_loss_d = 0.0for i, (text, images) in enumerate(dataloader):real_labels = torch.ones(images.size(0), 1)fake_labels = torch.zeros(images.size(0), 1)# Train Discriminatoroptimizer_d.zero_grad()real_outputs = discriminator(images)d_loss_real = criterion_gan(real_outputs, real_labels)generated_images = model(text)fake_outputs = discriminator(generated_images.detach())d_loss_fake = criterion_gan(fake_outputs, fake_labels)d_loss = (d_loss_real + d_loss_fake) / 2d_loss.backward()optimizer_d.step()# Train Generatoroptimizer_g.zero_grad()generated_images = model(text)g_outputs = discriminator(generated_images)g_loss_gan = criterion_gan(g_outputs, real_labels)g_loss_l1 = criterion_l1(generated_images, images)g_loss = g_loss_gan + 100 * g_loss_l1  # Weighted sum of GAN loss and L1 lossg_loss.backward()optimizer_g.step()running_loss_g += g_loss.item()running_loss_d += d_loss.item()print(f"Epoch {epoch + 1}, Generator Loss: {running_loss_g / len(dataloader)}, Discriminator Loss: {running_loss_d / len(dataloader)}")# 保存训练好的模型torch.save(model.state_dict(), config['model']['path'])# 图像生成
def generate_images(model, text_data, output_dir):for text in text_data:input_tensor = model.text_encoder([text])image = model.image_generator(input_tensor)image = image.squeeze(0).detach().cpu().numpy()image = (image * 127.5 + 127.5).astype('uint8')image = Image.fromarray(image.transpose(1, 2, 0))# 保存图像save_image(image, f"{output_dir}/{text}.png")# 图形用户界面
class TextToImageGUI:def __init__(self, root):self.root = rootself.root.title("文本生成图像")self.config = load_config('config.yaml')self.model = load_model(self.config['model']['path'], self.config['model']['text_encoder_model_name'])self.text_input = tk.Text(root, height=10, width=50)self.text_input.pack(pady=10)self.train_button = tk.Button(root, text="训练模型", command=self.train_model)self.train_button.pack(pady=10)self.generate_button = tk.Button(root, text="生成图像", command=self.generate_image)self.generate_button.pack(pady=10)self.image_label = tk.Label(root)self.image_label.pack(pady=10)def train_model(self):train_model(self.config)self.model = load_model(self.config['model']['path'], self.config['model']['text_encoder_model_name'])messagebox.showinfo("成功", "模型训练完成")def generate_image(self):text = self.text_input.get("1.0", tk.END).strip()if not text:messagebox.showwarning("警告", "请输入文本")returninput_tensor = self.model.text_encoder([text])image = self.model.image_generator(input_tensor)image = image.squeeze(0).detach().cpu().numpy()image = (image * 127.5 + 127.5).astype('uint8')image = Image.fromarray(image.transpose(1, 2, 0))# 显示图像img_tk = ImageTk.PhotoImage(image)self.image_label.config(image=img_tk)self.image_label.image = img_tk# 保存图像save_image(image, f"{self.config['data']['output_dir']}/{text}.png")messagebox.showinfo("成功", "图像已生成并保存")# 输出项目目录及所有文件
def list_files(startpath):for root, dirs, files in os.walk(startpath):level = root.replace(startpath, '').count(os.sep)indent = ' ' * 4 * (level)print('{}{}/'.format(indent, os.path.basename(root)))subindent = ' ' * 4 * (level + 1)for f in files:print('{}{}'.format(subindent, f))if __name__ == "__main__":config = load_config('config.yaml')# 输出项目目录及所有文件project_root = os.path.dirname(os.path.abspath(__file__))print("项目目录及所有文件:")list_files(project_root)# 加载模型model = load_model(config['model']['path'], config['model']['text_encoder_model_name'])# 加载文本数据text_data = load_text_data(config['data']['input_file'])# 生成图像generate_images(model, text_data, config['data']['output_dir'])# 启动图形用户界面root = tk.Tk()app = TextToImageGUI(root)root.mainloop()

6. 项目目录及所有文件输出
   在 main.py 中添加了一个 list_files 函数，用于输出项目目录及所有文件。这个函数会在程序启动时自动调用，输出当前项目的目录结构和所有文件。

7. 说明文档
   README.md
   在项目根目录下创建一个 README.md 文件，内容如下：

# 文本生成图像项目## 目录结构
text_to_image_project/
│
├── config.yaml
├── data/
│ ├── train_data.csv
│ └── test_data.txt
├── models/
│ └── text_to_image_model.pth
├── main.py
└── README.mdcode## 配置文件 `config.yaml`
```yaml
model:path: models/text_to_image_model.pthtext_encoder_model_name: bert-base-uncaseddata:input_file: data/test_data.txtoutput_dir: data/generated_imagesdataset_path: data/train_data.csvtraining:batch_size: 64learning_rate: 0.0002epochs: 100
训练数据文件
data/train_data.csv
csv
text,image_path
a beautiful sunset,data/images/sunset.jpg
a cute puppy,data/images/puppy.jpg
a red rose,data/images/rose.jpg
测试数据文件
data/test_data.txt
code
a beautiful sunset
a cute puppy
a red rose
运行项目
确保安装了所需的依赖库：bash
pip install torch torchvision transformers pillow tkinter
运行 main.py：bash
python main.py
功能
训练模型：点击“训练模型”按钮，开始训练模型。
生成图像：在文本框中输入文本，点击“生成图像”按钮，生成相应的图像并显示在界面上，同时保存到指定目录。
项目目录及所有文件
项目启动时会自动输出项目目录及所有文件。

希望这些改进和示例文件能帮助你更好地理解和运行项目。如果有任何问题或需要进一步的帮助，请随时告诉我！