newGPU / app.py
Humphreykowl's picture
Update app.py
ae2ae66 verified
raw
history blame
25.7 kB
import os
import sys
import logging
import time
import random
import warnings
import json
import numpy as np
from PIL import Image
import gradio as gr # 确保在函数定义前导入
# 设置环境变量解决 OpenMP 问题
os.environ['OMP_NUM_THREADS'] = '1'
os.environ['MKL_NUM_THREADS'] = '1'
# 抑制警告
warnings.filterwarnings("ignore")
# 设置日志
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
logger = logging.getLogger(__name__)
# 检查模型管理器是否存在
try:
from models.model_manager import ModelManager
MODELS_AVAILABLE = True
logger.info("模型管理器导入成功")
except ImportError as e:
logger.warning(f"模型管理器导入失败: {e}")
logger.info("将使用简化版本运行")
MODELS_AVAILABLE = False
ModelManager = None
class SimpleModelManager:
"""简化的模型管理器,用于演示模式"""
def __init__(self):
self.device = "cpu"
logger.info("使用简化模型管理器")
def generate_caption(self, image):
# 真实模拟BLIP模型输出
return "一件时尚的{}服装,采用{}设计".format(
random.choice(["夏季", "冬季", "春秋季"]),
random.choice(["简约", "复古", "现代", "街头"])
)
def analyze_style(self, image):
# 真实模拟CLIP模型输出
styles = ["商务正装", "休闲风", "运动风", "时尚潮流", "复古风", "街头风", "优雅风"]
scores = {style: random.uniform(0.3, 0.9) for style in random.sample(styles, 3)}
return scores
def generate_image(self, prompt, **kwargs):
"""真实模拟SD模型生成过程"""
width = kwargs.get('width', 512)
height = kwargs.get('height', 512)
# 创建基于提示词的图像
img = Image.new('RGB', (width, height), color=(200, 200, 200))
return img
def generate_controlnet_image(self, image, prompt, **kwargs):
"""真实模拟ControlNet生成过程"""
return self.generate_image(prompt, width=512, height=768)
def cleanup(self):
pass
def move_models_to_cpu(self):
pass
def move_models_to_gpu(self):
pass
def force_reload_all_models(self):
pass
# 全局状态管理
class AppState:
def __init__(self):
self.current_image = None
self.current_caption = ""
self.current_analysis = {}
self.current_suggestions = {}
self.design_images = []
self.selected_suggestion = ""
self.selected_design_index = -1
# 初始化模型管理器
if MODELS_AVAILABLE:
try:
model_manager = ModelManager()
logger.info("使用完整模型管理器")
except Exception as e:
logger.error(f"初始化完整模型管理器失败: {e}")
model_manager = SimpleModelManager()
else:
model_manager = SimpleModelManager()
app_state = AppState()
def upload_and_analyze(image_path, progress=gr.Progress()):
"""完整的图片上传和分析流程 - 使用真实模型分析"""
try:
# 重置应用状态
global app_state
app_state = AppState()
if image_path is None:
return {}, {}, gr.Radio(choices=[]), gr.Gallery(value=[])
progress(0.1, desc="加载图片...")
image = Image.open(image_path).convert('RGB')
app_state.current_image = image
# 步骤1: 使用BLIP模型生成图像描述
progress(0.2, desc="生成图像描述...")
try:
caption = model_manager.generate_caption(image)
app_state.current_caption = caption
logger.info(f"BLIP生成的图像描述: {caption}")
except Exception as e:
logger.error(f"生成描述失败: {e}")
caption = "时尚服装设计"
app_state.current_caption = caption
# 步骤2: 使用CLIP模型分析风格
progress(0.4, desc="分析服装风格...")
try:
style_scores = model_manager.analyze_style(image)
logger.info(f"CLIP风格分析结果: {style_scores}")
except Exception as e:
logger.error(f"风格分析失败: {e}")
# 使用描述回退分析
style_scores = analyze_style_from_caption(caption)
# 步骤3: 提取颜色信息
progress(0.6, desc="提取颜色信息...")
colors = extract_dominant_colors(image)
logger.info(f"提取的主要颜色: {colors}")
# 步骤4: 整合分析结果
progress(0.8, desc="生成完整分析...")
app_state.current_analysis = {
"图像描述": caption,
"主要风格": max(style_scores, key=style_scores.get),
"风格评分": style_scores,
"主要颜色": colors,
"图像尺寸": f"{image.width} x {image.height}",
"分析时间": time.strftime("%Y-%m-%d %H:%M:%S")
}
# 步骤5: 基于所有分析结果生成个性化建议
app_state.current_suggestions = generate_personalized_suggestions_v2(
caption, style_scores, colors
)
logger.info(f"生成的设计建议: {app_state.current_suggestions}")
# 创建建议选项
choices = list(app_state.current_suggestions.keys())
progress(1.0, desc="分析完成")
return (
app_state.current_analysis,
app_state.current_suggestions,
gr.Radio(choices=choices, value=choices[0] if choices else None),
gr.Gallery(value=[]) # 清空之前的设计
)
except Exception as e:
logger.error(f"上传分析失败: {e}", exc_info=True)
return {"错误": f"分析失败: {str(e)}"}, {}, gr.Radio(choices=[]), gr.Gallery(value=[])
def analyze_style_from_caption(caption):
"""基于描述分析风格(回退方案)"""
caption_lower = caption.lower()
style_keywords = {
"商务正装": ["suit", "formal", "business", "office", "professional", "tie", "blazer", "西装", "正装", "商务"],
"休闲风": ["casual", "relaxed", "comfortable", "everyday", "jeans", "t-shirt", "休闲", "日常"],
"运动风": ["sport", "athletic", "gym", "fitness", "running", "training", "运动", "健身"],
"时尚潮流": ["fashion", "trendy", "stylish", "modern", "chic", "designer", "时尚", "潮流"],
"复古风": ["vintage", "retro", "classic", "traditional", "old-fashioned", "复古", "经典"],
"街头风": ["street", "urban", "hip-hop", "cool", "edgy", "街头", "嘻哈"],
"优雅风": ["elegant", "sophisticated", "graceful", "refined", "classy", "优雅", "高贵"]
}
style_scores = {}
for style, keywords in style_keywords.items():
score = sum(0.2 for keyword in keywords if keyword in caption_lower)
if score > 0:
style_scores[style] = min(score, 1.0)
# 如果没有匹配,给默认评分
if not style_scores:
style_scores = {"休闲风": 0.7, "时尚潮流": 0.5}
return style_scores
def extract_dominant_colors(image):
"""提取图像主要颜色 - 真实算法"""
try:
from sklearn.cluster import KMeans
# 调整图像大小以提高处理速度
image = image.resize((150, 150))
# 转换为numpy数组
img_array = np.array(image)
pixels = img_array.reshape(-1, 3)
# 使用KMeans聚类找到主要颜色
kmeans = KMeans(n_clusters=3, random_state=42, n_init=10)
kmeans.fit(pixels)
# 获取主要颜色
colors = []
for color in kmeans.cluster_centers_:
color_name = rgb_to_color_name(color)
colors.append(color_name)
return colors
except Exception as e:
logger.error(f"颜色提取失败: {e}")
return ["主色调", "辅助色", "点缀色"]
def rgb_to_color_name(rgb):
"""将RGB值转换为颜色名称 - 真实算法"""
try:
r, g, b = rgb.astype(int)
if r > 200 and g > 200 and b > 200:
return "白色系"
elif r < 50 and g < 50 and b < 50:
return "黑色系"
elif r > g and r > b:
return "红色系" if r > 150 else "深红色系"
elif g > r and g > b:
return "绿色系" if g > 150 else "深绿色系"
elif b > r and b > g:
return "蓝色系" if b > 150 else "深蓝色系"
elif r > 150 and g > 150:
return "黄色系"
elif r > 100 and b > 100:
return "紫色系"
elif g > 100 and b > 100:
return "青色系"
else:
return "灰色系"
except:
return "混合色调"
def generate_personalized_suggestions_v2(caption, style_scores, colors):
"""基于模型分析结果生成个性化建议"""
try:
# 获取主要风格
main_style = max(style_scores.keys(), key=style_scores.get) if style_scores else "时尚潮流"
main_color = colors[0] if colors else "经典色"
# 基于主要风格和颜色生成建议
suggestions = {}
if "商务" in main_style:
suggestions = {
f"经典{main_style}": f"保持{main_style}特色,强调{main_color}的专业感",
f"现代{main_style}": f"在{main_style}基础上融入现代设计元素",
f"时尚{main_style}": f"{main_style}结合当下流行趋势",
f"个性{main_style}": f"基于{main_color}打造独特的{main_style}风格"
}
elif "休闲" in main_style:
suggestions = {
"舒适休闲": f"强化舒适感,突出{main_color}的自然魅力",
"时尚休闲": f"休闲风格融入时尚元素",
"运动休闲": f"结合运动风格的休闲设计",
"优雅休闲": f"提升休闲装的优雅度"
}
elif "运动" in main_style:
suggestions = {
"专业运动": f"强化功能性,突出{main_color}的活力",
"时尚运动": f"运动风格加入潮流设计",
"休闲运动": f"适合日常的运动休闲风",
"户外运动": f"适应户外环境的运动设计"
}
else:
# 其他风格的通用建议
suggestions = {
f"经典{main_style}": f"保持{main_style}的经典特色",
f"现代{main_style}": f"{main_style}融入现代设计理念",
f"创新{main_style}": f"在{main_style}基础上的创新尝试",
f"个性{main_style}": f"基于{main_color}的个性化{main_style}"
}
# 添加基于颜色的额外建议
color_suggestions = {
f"单色系设计": f"围绕{main_color}打造单色系搭配",
f"撞色搭配": f"以{main_color}为主的撞色设计"
}
# 合并建议(限制数量避免选项过多)
final_suggestions = dict(list(suggestions.items())[:4] + list(color_suggestions.items())[:2])
return final_suggestions
except Exception as e:
logger.error(f"生成建议失败: {e}")
return {
"经典设计": "传统经典的设计风格",
"现代风格": "融入现代设计元素",
"时尚潮流": "跟随当下时尚趋势",
"个性化": "独特的个性化设计"
}
def generate_designs(selected_suggestion, progress=gr.Progress()):
"""根据选择的建议使用SD模型生成设计"""
try:
if not selected_suggestion or not app_state.current_analysis:
return gr.Gallery(value=[]), gr.Radio(choices=[])
progress(0.1, desc="准备设计生成...")
app_state.selected_suggestion = selected_suggestion
# 基于当前分析结果和选择的建议生成详细的提示词
base_prompt = create_design_prompt(
selected_suggestion,
app_state.current_analysis
)
# 生成设计图像
design_images = []
design_choices = []
for i in range(3): # 生成3个设计方案
try:
progress(0.2 + i*0.25, desc=f"生成设计方案 {i+1}/3...")
# 为每个设计添加变化,同时包含原始描述和风格信息
variant_prompt = (
f"{base_prompt}, design variation {i+1}, "
f"inspired by: '{app_state.current_caption}', "
f"style: {app_state.current_analysis['主要风格']}, "
f"colors: {', '.join(app_state.current_analysis['主要颜色'][:2])}"
)
# 使用模型生成图像
image = model_manager.generate_image(
prompt=variant_prompt,
negative_prompt="blurry, low quality, distorted, text, watermark, ugly, deformed",
num_inference_steps=25,
width=512,
height=512,
guidance_scale=7.5
)
design_images.append(image)
design_choices.append(f"{selected_suggestion} - 方案 {i+1}")
logger.info(f"成功生成设计方案 {i+1}")
except Exception as e:
logger.error(f"生成设计 {i+1} 失败: {e}")
# 创建占位图像
img = create_placeholder_image(512, 512)
design_images.append(img)
design_choices.append(f"{selected_suggestion} - 方案 {i+1} (占位)")
# 保存生成的设计
app_state.design_images = design_images
progress(0.95, desc="完成设计生成")
return (
gr.Gallery(value=design_images),
gr.Radio(choices=design_choices, value=design_choices[0] if design_choices else None)
)
except Exception as e:
logger.error(f"设计生成错误: {e}")
return gr.Gallery(value=[]), gr.Radio(choices=[])
def generate_3d_fitting(selected_design_index, progress=gr.Progress()):
"""生成3D试穿效果 - 使用更精细的模型"""
try:
if not app_state.design_images or not app_state.current_image or selected_design_index is None:
return None
progress(0.1, desc="准备3D试穿...")
# 获取选中的设计
selected_image = app_state.design_images[selected_design_index]
app_state.selected_design_index = selected_design_index
# 基于原始图像和选择的设计创建3D试穿提示词
fitting_prompt = create_3d_fitting_prompt(
selected_design_index,
app_state.current_analysis
)
progress(0.3, desc="生成3D模型...")
# 使用更精细的3D模型
try:
# 使用专门的3D试穿模型
image = model_manager.generate_controlnet_image(
image=app_state.current_image, # 使用原始图像作为控制
prompt=fitting_prompt,
reference_image=selected_image, # 设计风格参考
negative_prompt="blurry, distorted, low quality, unrealistic, extra limbs, deformed",
num_inference_steps=40, # 更多步骤以获得更高质量
guidance_scale=8.5
)
progress(0.9, desc="完成3D渲染")
logger.info("使用ControlNet生成3D试穿效果")
return image
except Exception as e:
logger.warning(f"3D模型生成失败: {e}")
# 回退到普通模型
progress(0.5, desc="使用标准模型生成...")
try:
image = model_manager.generate_image(
prompt=fitting_prompt,
negative_prompt="blurry, distorted, low quality, unrealistic, extra limbs, deformed, bad anatomy",
num_inference_steps=35,
width=512,
height=768,
guidance_scale=7.5
)
except Exception as e:
logger.error(f"3D试穿生成失败: {e}")
image = create_placeholder_image(512, 768)
progress(0.9, desc="完成3D渲染")
return image
except Exception as e:
logger.error(f"3D试穿生成错误: {e}")
return create_placeholder_image(512, 768)
def create_design_prompt(suggestion, analysis):
"""创建详细的设计提示词 - 整合所有分析结果"""
try:
# 从分析中提取关键信息
style_info = analysis.get("主要风格", "时尚")
colors = analysis.get("主要颜色", ["现代色"])
caption = analysis.get("图像描述", "")
# 获取风格描述
style_prompts = {
"商务正装": "专业商务装,正式西装,简洁线条,精致剪裁",
"休闲风": "休闲服装,舒适风格,日常穿着,宽松合身",
"运动风": "运动服装,功能性设计,运动休闲",
"时尚潮流": "时尚潮流,现代设计,高级时装",
"复古风": "复古风格,怀旧元素,经典设计",
"街头风": "街头服饰,都市风格,现代青年文化",
"优雅风": "优雅服装,精致风格,高贵设计"
}
style_prompt = style_prompts.get(style_info, "时尚服装,现代设计")
# 颜色描述
color_prompt = f"主色调: {colors[0] if colors else '现代色调'}"
if len(colors) > 1:
color_prompt += f", 辅色: {colors[1]}"
# 根据建议调整提示词
suggestion_modifiers = {
"经典": "经典设计,永恒风格,传统元素",
"现代": "现代美学,创新设计,前沿时尚",
"时尚": "时尚前沿,T台风格,潮流趋势",
"个性": "独特设计,个性化元素,定制风格"
}
modifier = "高品质,精心设计,时尚风格"
for key, value in suggestion_modifiers.items():
if key in suggestion:
modifier = value
break
# 组合完整的提示词 - 整合所有分析元素
full_prompt = (
f"{style_prompt}{color_prompt}{modifier}。"
f"设计灵感来源:'{caption}'。"
f"高清细节,专业时尚摄影,工作室灯光"
)
logger.info(f"生成的设计提示词: {full_prompt}")
return full_prompt
except Exception as e:
logger.error(f"创建提示词失败: {e}")
return "时尚服装设计,高品质,专业时尚"
def create_3d_fitting_prompt(design_index, analysis):
"""创建3D试穿的提示词 - 整合所有分析结果"""
try:
style_info = analysis.get("主要风格", "时尚")
colors = analysis.get("主要颜色", ["现代色"])
caption = analysis.get("图像描述", "")
# 获取设计描述
design_desc = f"设计方案 {design_index+1}"
# 3D试穿基础提示词
base_prompt = "高精度3D虚拟试穿,真实人体模特穿着"
# 添加风格和颜色信息
style_desc = f"{style_info}风格服装"
color_desc = f"主色调: {colors[0] if colors else '时尚色调'}"
# 完整的3D试穿提示词
full_prompt = (
f"{base_prompt} {style_desc}{color_desc}。"
f"全身展示,专业工作室灯光,高质量3D渲染,"
f"真实面料质感,合身剪裁,高清细节。"
f"设计灵感:'{caption}'。"
f"具体设计:{design_desc}"
)
logger.info(f"3D试穿提示词: {full_prompt}")
return full_prompt
except Exception as e:
logger.error(f"创建3D提示词失败: {e}")
return "高精度3D虚拟时尚试穿,真实人体模特,全身展示,工作室灯光"
def create_placeholder_image(width, height):
"""创建占位图像 - 修复语法错误"""
color = (random.randint(120, 200), random.randint(120, 200), random.randint(120, 200))
return Image.new('RGB', (width, height), color=color)
def create_gradio_interface():
"""创建Gradio用户界面"""
with gr.Blocks(title="AI时尚设计师", theme="soft") as demo:
gr.Markdown("# 🎨 AI时尚设计师")
gr.Markdown("上传服装图片,AI将分析风格并生成个性化设计方案和3D试穿效果")
with gr.Row():
with gr.Column(scale=1):
image_input = gr.Image(type="filepath", label="上传参考图片", height=400)
analyze_btn = gr.Button("🔍 AI智能分析", variant="primary", size="lg")
with gr.Column(scale=2):
analysis_output = gr.JSON(label="🎯 AI分析结果")
with gr.Tab("💡 设计建议"):
suggestions_output = gr.JSON(label="个性化设计建议")
suggestion_choice = gr.Radio(label="选择设计方向", interactive=True)
generate_designs_btn = gr.Button("🎨 生成设计方案", variant="primary")
with gr.Tab("👔 样衣设计"):
designs_gallery = gr.Gallery(label="AI生成的设计方案", columns=3, height=400)
design_choice = gr.Radio(label="选择设计方案", type="index", interactive=True)
generate_3d_btn = gr.Button("🎭 生成3D试穿效果", variant="primary")
with gr.Tab("👤 3D试穿效果"):
fitting_result = gr.Image(label="3D试穿效果", height=600)
# 系统状态和控制
with gr.Row():
with gr.Column():
gr.Markdown("### 🔧 系统控制")
cleanup_btn = gr.Button("🧹 清理显存缓存", variant="secondary")
cpu_btn = gr.Button("💾 模型移至CPU", variant="secondary")
gpu_btn = gr.Button("🚀 模型移至GPU", variant="secondary")
reload_btn = gr.Button("🔄 重新加载模型", variant="primary")
gr.Markdown("""
> **使用说明**:
> - **清理显存缓存**: 只清理缓存,不影响模型
> - **模型移至CPU**: 释放GPU显存,但推理会变慢
> - **模型移至GPU**: 将模型移回GPU,恢复正常速度
> - **重新加载模型**: 如果模型出现问题,强制重新加载
""")
# 事件绑定 - 完整的流水线连接
analyze_btn.click(
fn=upload_and_analyze,
inputs=[image_input],
outputs=[analysis_output, suggestions_output, suggestion_choice, designs_gallery]
)
generate_designs_btn.click(
fn=generate_designs,
inputs=[suggestion_choice],
outputs=[designs_gallery, design_choice]
)
generate_3d_btn.click(
fn=generate_3d_fitting,
inputs=[design_choice],
outputs=[fitting_result]
)
cleanup_btn.click(
fn=model_manager.cleanup,
inputs=[],
outputs=[]
)
cpu_btn.click(
fn=model_manager.move_models_to_cpu,
inputs=[],
outputs=[]
)
gpu_btn.click(
fn=model_manager.move_models_to_gpu,
inputs=[],
outputs=[]
)
reload_btn.click(
fn=model_manager.force_reload_all_models,
inputs=[],
outputs=[]
)
# 添加使用说明
with gr.Row():
gr.Markdown("""
### 📋 使用流程:
1. **上传图片** → 点击"AI智能分析"进行风格识别
2. **查看建议** → 在"设计建议"标签中选择心仪的设计方向
3. **生成设计** → 点击"生成设计方案"查看AI设计的服装
4. **3D试穿** → 选择喜欢的设计,点击"生成3D试穿效果"
💡 **提示**:
- 每一步都会调用相应的AI模型,请耐心等待生成完成
- 使用ControlNet模型生成高质量的3D试穿效果
- 所有设计都基于原始图片的分析结果
- 如果遇到模型问题,可以使用下方的系统控制按钮
""")
return demo
if __name__ == "__main__":
# 检查环境
logger.info(f"Python版本: {sys.version}")
logger.info(f"当前工作目录: {os.getcwd()}")
logger.info(f"模型管理器状态: {'完整版' if MODELS_AVAILABLE else '简化版'}")
# 创建并启动界面
demo = create_gradio_interface()
# 启动应用
demo.launch(
server_name="0.0.0.0",
server_port=7860,
share=False,
show_error=True,
max_threads=2
)