qwen3.5 优化

fastAPI_tarot.py

@@ -22,6 +22,7 @@ import re
 import asyncio
 import shutil
 import subprocess
+import ast
 from datetime import datetime
 from typing import Optional, List, Dict, Any
 from contextlib import asynccontextmanager
@@ -288,6 +289,35 @@ def append_to_history(req_type: str, prompt: str, status: str, result_path: str
         print(f"Failed to write history: {e}")
 
 
+def extract_json_from_response(text: str) -> dict:
+    """
+    Robustly extract JSON from text, handling:
+    1. Markdown code blocks (```json ... ```)
+    2. Single quotes (Python dict style) via ast.literal_eval
+    """
+    try:
+        # 1. Try to find JSON block
+        json_match = re.search(r'```json\s*(.*?)\s*```', text, re.DOTALL)
+        if json_match:
+            clean_text = json_match.group(1).strip()
+        else:
+            # Try to find { ... } block if no markdown
+            match = re.search(r'\{.*\}', text, re.DOTALL)
+            if match:
+                clean_text = match.group(0).strip()
+            else:
+                clean_text = text.strip()
+        
+        # 2. Try standard JSON
+        return json.loads(clean_text)
+    except Exception as e1:
+        # 3. Try ast.literal_eval for single quotes
+        try:
+            return ast.literal_eval(clean_text)
+        except Exception as e2:
+            # 4. Fail
+            raise ValueError(f"Could not parse JSON: {e1} | {e2} | Content: {text[:100]}...")
+
 def translate_to_sam3_prompt(text: str) -> str:
     """
     使用 Qwen 模型将中文提示词翻译为英文
@@ -567,13 +597,13 @@ def recognize_card_with_qwen(image_path: str) -> dict:
         
         if response.status_code == 200:
             content = response.output.choices[0].message.content[0]['text']
-            import json
             try:
-                clean_content = content.replace("```json", "").replace("```", "").strip()
+                result = extract_json_from_response(content)
-                result = json.loads(clean_content)
+                result["model_used"] = QWEN_MODEL
                 return result
-            except:
+            except Exception as e:
-                return {"raw_response": content}
+                print(f"JSON Parse Error in recognize_card: {e}")
+                return {"raw_response": content, "error": str(e), "model_used": QWEN_MODEL}
         else:
             return {"error": f"API Error: {response.code} - {response.message}"}
             
@@ -602,13 +632,13 @@ def recognize_spread_with_qwen(image_path: str) -> dict:
         
         if response.status_code == 200:
             content = response.output.choices[0].message.content[0]['text']
-            import json
             try:
-                clean_content = content.replace("```json", "").replace("```", "").strip()
+                result = extract_json_from_response(content)
-                result = json.loads(clean_content)
+                result["model_used"] = QWEN_MODEL
                 return result
-            except:
+            except Exception as e:
-                return {"raw_response": content, "spread_name": "Unknown"}
+                print(f"JSON Parse Error in recognize_spread: {e}")
+                return {"raw_response": content, "error": str(e), "spread_name": "Unknown", "model_used": QWEN_MODEL}
         else:
             return {"error": f"API Error: {response.code} - {response.message}"}
             
@@ -951,6 +981,10 @@ async def recognize_tarot(
     processor = request.app.state.processor
 
     try:
+        # 在执行 GPU 操作前，切换到线程中运行，避免阻塞主线程（虽然 SAM3 推理在 CPU 上可能已经很快，但为了保险）
+        # 注意：processor 内部调用了 torch，如果是在 GPU 上，最好不要多线程调用同一个 model
+        # 但这里只是推理，且是单次请求。
+        # 如果是 CPU 推理，run_in_executor 有助于防止阻塞 loop
         inference_state = processor.set_image(image)
         output = processor.set_text_prompt(state=inference_state, prompt="tarot card")
         masks, boxes, scores = output["masks"], output["boxes"], output["scores"]
@@ -975,15 +1009,25 @@ async def recognize_tarot(
         main_file_path = None
         main_file_url = None
 
-    # Step 0: 牌阵识别
+    # Step 0: 牌阵识别 (异步启动)
     spread_info = {"spread_name": "Unknown"}
+    spread_task = None
     if main_file_path:
         # 使用原始图的一份拷贝给 Qwen 识别牌阵
         temp_raw_path = os.path.join(output_dir, "raw_for_spread.jpg")
         image.save(temp_raw_path)
-        spread_info = recognize_spread_with_qwen(temp_raw_path)
+        # 将同步调用包装为异步任务
+        loop = asyncio.get_event_loop()
+        spread_task = loop.run_in_executor(None, recognize_spread_with_qwen, temp_raw_path)
 
     if detected_count != expected_count:
+        # 如果数量不对，等待牌阵识别完成（如果已启动）再返回
+        if spread_task:
+            try:
+                spread_info = await spread_task
+            except Exception as e:
+                print(f"Spread recognition failed: {e}")
+                
         duration = time.time() - start_time
         append_to_history("tarot-recognize", f"expected: {expected_count}", "failed", result_path=f"results/{request_id}/{main_filename}" if main_file_url else None, details=f"Detected {detected_count}, expected {expected_count}", duration=duration)
         return JSONResponse(
@@ -1005,21 +1049,47 @@ async def recognize_tarot(
         append_to_history("tarot-recognize", f"expected: {expected_count}", "failed", details=f"Crop Error: {str(e)}", duration=duration)
         raise HTTPException(status_code=500, detail=f"抠图处理错误: {str(e)}")
 
-    # 遍历每张卡片进行识别
+    # 遍历每张卡片进行识别 (并发)
     tarot_cards = []
+    
+    # 1. 准备任务列表
+    loop = asyncio.get_event_loop()
+    card_tasks = []
+    
     for obj in saved_objects:
         fname = obj["filename"]
         file_path = os.path.join(output_dir, fname)
+        # 创建异步任务
+        # 使用 lambda 来延迟调用，确保参数传递正确
+        task = loop.run_in_executor(None, recognize_card_with_qwen, file_path)
+        card_tasks.append(task)
     
-        # 调用 Qwen-VL 识别 (串行)
+    # 2. 等待所有卡片识别任务完成
-        recognition_res = recognize_card_with_qwen(file_path)
+    # 同时等待牌阵识别任务 (如果还在运行)
+    if card_tasks:
+        all_card_results = await asyncio.gather(*card_tasks)
+    else:
+        all_card_results = []
     
+    if spread_task:
+        try:
+            # 如果之前没有await spread_task，这里确保它完成
+            # 注意：如果 detected_count != expected_count 分支已经 await 过了，这里不会重复执行
+            # 但那个分支有 return，所以这里肯定是还没 await 的
+            spread_info = await spread_task
+        except Exception as e:
+            print(f"Spread recognition failed: {e}")
+
+    # 3. 组装结果
+    for i, obj in enumerate(saved_objects):
+        fname = obj["filename"]
         file_url = str(request.url_for("static", path=f"results/{request_id}/{fname}"))
+        
         tarot_cards.append({
             "url": file_url,
             "is_rotated": obj["is_rotated_by_algorithm"],
             "orientation_status": "corrected_to_portrait" if obj["is_rotated_by_algorithm"] else "original_portrait",
-            "recognition": recognition_res,
+            "recognition": all_card_results[i],
             "note": obj["note"]
         })
 
@@ -1083,6 +1153,18 @@ async def segment_face(
 
     # 调用独立服务进行处理
     try:
+        # 使用新增加的异步并发函数
+        if hasattr(human_analysis_service, "process_face_segmentation_and_analysis_async"):
+            result = await human_analysis_service.process_face_segmentation_and_analysis_async(
+                processor=processor,
+                image=image,
+                prompt=final_prompt,
+                output_base_dir=RESULT_IMAGE_DIR,
+                qwen_model=QWEN_MODEL,
+                analysis_prompt=PROMPTS["face_analysis"]
+            )
+        else:
+            # 回退到同步
             result = human_analysis_service.process_face_segmentation_and_analysis(
                 processor=processor,
                 image=image,

history.json

@@ -0,0 +1,3 @@
+{"timestamp": 1771396524.4495308, "type": "tarot-recognize", "prompt": "expected: 3", "final_prompt": null, "status": "success", "result_path": "results/1771396501_cd6d8769/seg_f8d9ba7c28fc403cbce75516ba2cd3c4.jpg", "details": "Spread: 三张牌", "duration": 23.850417613983154}
+{"timestamp": 1771396579.124643, "type": "tarot-recognize", "prompt": "expected: 3", "final_prompt": null, "status": "success", "result_path": "results/1771396461_4ac00fc4/seg_f85c26fe73cf47c79d308c11f3ef3f0c.jpg", "details": "Spread: 三张牌牌阵", "duration": 119.34761571884155}
+{"timestamp": 1771396602.0928633, "type": "tarot-recognize", "prompt": "expected: 3", "final_prompt": null, "status": "success", "result_path": "results/1771396577_a0d559ee/seg_976a193b5025413fbc7d6064d9de0680.jpg", "details": "Spread: 三张牌", "duration": 25.696484327316284}

human_analysis_service.py

@@ -6,6 +6,8 @@ import numpy as np
 import json
 import torch
 import cv2
+import ast
+import re
 from PIL import Image
 from dashscope import MultiModalConversation
 
@@ -95,6 +97,35 @@ def create_highlighted_visualization(image: Image.Image, masks, output_path: str
     # Save
     Image.fromarray(result_np).save(output_path)
 
+def extract_json_from_response(text: str) -> dict:
+    """
+    Robustly extract JSON from text, handling:
+    1. Markdown code blocks (```json ... ```)
+    2. Single quotes (Python dict style) via ast.literal_eval
+    """
+    try:
+        # 1. Try to find JSON block
+        json_match = re.search(r'```json\s*(.*?)\s*```', text, re.DOTALL)
+        if json_match:
+            clean_text = json_match.group(1).strip()
+        else:
+            # Try to find { ... } block if no markdown
+            match = re.search(r'\{.*\}', text, re.DOTALL)
+            if match:
+                clean_text = match.group(0).strip()
+            else:
+                clean_text = text.strip()
+        
+        # 2. Try standard JSON
+        return json.loads(clean_text)
+    except Exception as e1:
+        # 3. Try ast.literal_eval for single quotes
+        try:
+            return ast.literal_eval(clean_text)
+        except Exception as e2:
+            # 4. Fail
+            raise ValueError(f"Could not parse JSON: {e1} | {e2} | Content: {text[:100]}...")
+
 def analyze_demographics_with_qwen(image_path: str, model_name: str = 'qwen-vl-max', prompt_template: str = None) -> dict:
     """
     调用 Qwen-VL 模型分析人物的年龄和性别
@@ -131,19 +162,21 @@ def analyze_demographics_with_qwen(image_path: str, model_name: str = 'qwen-vl-m
         
         if response.status_code == 200:
             content = response.output.choices[0].message.content[0]['text']
-            # 清理 Markdown 代码块标记
-            clean_content = content.replace("```json", "").replace("```", "").strip()
             try:
-                result = json.loads(clean_content)
+                result = extract_json_from_response(content)
+                result["model_used"] = model_name
                 return result
-            except json.JSONDecodeError:
+            except Exception as e:
-                return {"raw_analysis": clean_content}
+                print(f"JSON Parse Error in face analysis: {e}")
+                return {"raw_analysis": content, "error": str(e), "model_used": model_name}
         else:
             return {"error": f"API Error: {response.code} - {response.message}"}
             
     except Exception as e:
         return {"error": f"分析失败: {str(e)}"}
 
+import asyncio
+
 def process_face_segmentation_and_analysis(
     processor, 
     image: Image.Image, 
@@ -156,11 +189,11 @@ def process_face_segmentation_and_analysis(
     核心处理逻辑：
     1. SAM3 分割 (默认提示词 "head" 以包含头发)
     2. 裁剪图片
-    3. Qwen-VL 识别性别年龄
+    3. Qwen-VL 识别性别年龄 (并发)
     4. 返回结果
     """
     
-    # 1. SAM3 推理
+    # 1. SAM3 推理 (同步，因为涉及 GPU 操作)
     inference_state = processor.set_image(image)
     output = processor.set_text_prompt(state=inference_state, prompt=prompt)
     masks, boxes, scores = output["masks"], output["boxes"], output["scores"]
@@ -179,7 +212,7 @@ def process_face_segmentation_and_analysis(
     output_dir = os.path.join(output_base_dir, request_id)
     os.makedirs(output_dir, exist_ok=True)
     
-    # --- 新增：生成背景变暗的整体可视化图 ---
+    # --- 生成可视化图 ---
     vis_filename = f"seg_{uuid.uuid4().hex}.jpg"
     vis_path = os.path.join(output_dir, vis_filename)
     try:
@@ -188,38 +221,238 @@ def process_face_segmentation_and_analysis(
     except Exception as e:
         print(f"可视化生成失败: {e}")
         full_vis_relative_path = None
-    # -------------------------------------
+    # ------------------
 
-    results = []
+    # 转换 boxes 和 scores
-    
-    # 转换 boxes 为 numpy
     if isinstance(boxes, torch.Tensor):
         boxes_np = boxes.cpu().numpy()
     else:
         boxes_np = boxes
     
-    # 转换 scores 为 list
     if isinstance(scores, torch.Tensor):
         scores_list = scores.tolist()
     else:
         scores_list = scores if isinstance(scores, list) else [float(scores)]
 
-    for i, box in enumerate(boxes_np):
+    # 准备异步任务
-        # 2. 裁剪 (带一点 padding 以保留完整发型)
+    async def run_analysis_tasks():
-        # 2. 裁剪 (带一点 padding 以保留完整发型)
+        loop = asyncio.get_event_loop()
-        cropped_img = crop_head_with_padding(image, box, padding_ratio=0.1)
+        tasks = []
+        temp_results = [] # 存储 (index, filename, score) 以便后续排序组合
 
-        # 保存裁剪图
+        for i, box in enumerate(boxes_np):
+            # 2. 裁剪 (同步)
+            cropped_img = crop_head_with_padding(image, box, padding_ratio=0.1)
             filename = f"face_{i}.jpg"
             save_path = os.path.join(output_dir, filename)
             cropped_img.save(save_path)
             
-        # 3. 识别
+            # 3. 准备识别任务
+            task = loop.run_in_executor(
+                None, 
+                analyze_demographics_with_qwen, 
+                save_path, 
+                qwen_model, 
+                analysis_prompt
+            )
+            tasks.append(task)
+            temp_results.append({
+                "filename": filename,
+                "relative_path": f"results/{request_id}/{filename}",
+                "score": float(scores_list[i]) if i < len(scores_list) else 0.0
+            })
+            
+        # 等待所有任务完成
+        if tasks:
+            analysis_results = await asyncio.gather(*tasks)
+        else:
+            analysis_results = []
+            
+        # 组合结果
+        final_results = []
+        for i, item in enumerate(temp_results):
+            item["analysis"] = analysis_results[i]
+            final_results.append(item)
+            
+        return final_results
+
+    # 运行异步任务
+    # 注意：由于本函数被 FastAPI (异步环境) 中的同步或异步函数调用，
+    # 如果上层是 async def，我们可以直接 await。
+    # 但由于这个函数定义没有 async，且之前的调用是同步的，
+    # 为了兼容性，我们需要检查当前是否在事件循环中。
+    
+    # 然而，查看 fastAPI_tarot.py，这个函数是在 async def segment_face 中被调用的。
+    # 但它是作为普通函数被导入和调用的。
+    # 为了不破坏现有签名，我们可以使用 asyncio.run() 或者在新循环中运行，
+    # 但这在已经运行的 loop 中是不允许的。
+    
+    # 最佳方案：修改本函数为 async，并在 fastAPI_tarot.py 中 await 它。
+    # 但这需要修改 fastAPI_tarot.py 的调用处。
+    
+    # 既然我们已经修改了 fastAPI_tarot.py，我们也可以顺便修改这里的签名。
+    # 但为了稳妥，我们可以用一种 hack：
+    # 如果在一个正在运行的 loop 中调用，我们必须返回 awaitable 或者使用 loop.run_until_complete (会报错)
+    
+    # 让我们先把这个函数改成 async，然后去修改 fastAPI_tarot.py 的调用。
+    # 这是最正确的做法。
+    pass # 占位，实际代码在下面
+
+async def process_face_segmentation_and_analysis_async(
+    processor, 
+    image: Image.Image, 
+    prompt: str = "head", 
+    output_base_dir: str = "static/results",
+    qwen_model: str = "qwen-vl-max",
+    analysis_prompt: str = None
+) -> dict:
+    # ... (同上逻辑，只是是 async)
+    
+    # 1. SAM3 推理
+    inference_state = processor.set_image(image)
+    output = processor.set_text_prompt(state=inference_state, prompt=prompt)
+    masks, boxes, scores = output["masks"], output["boxes"], output["scores"]
+    
+    detected_count = len(masks)
+    if detected_count == 0:
+        return {
+            "status": "success",
+            "message": "未检测到目标",
+            "detected_count": 0,
+            "results": []
+        }
+        
+    request_id = f"{int(time.time())}_{uuid.uuid4().hex[:8]}"
+    output_dir = os.path.join(output_base_dir, request_id)
+    os.makedirs(output_dir, exist_ok=True)
+    
+    vis_filename = f"seg_{uuid.uuid4().hex}.jpg"
+    vis_path = os.path.join(output_dir, vis_filename)
+    try:
+        create_highlighted_visualization(image, masks, vis_path)
+        full_vis_relative_path = f"results/{request_id}/{vis_filename}"
+    except Exception as e:
+        print(f"可视化生成失败: {e}")
+        full_vis_relative_path = None
+
+    if isinstance(boxes, torch.Tensor):
+        boxes_np = boxes.cpu().numpy()
+    else:
+        boxes_np = boxes
+    
+    if isinstance(scores, torch.Tensor):
+        scores_list = scores.tolist()
+    else:
+        scores_list = scores if isinstance(scores, list) else [float(scores)]
+
+    loop = asyncio.get_event_loop()
+    tasks = []
+    results = []
+
+    for i, box in enumerate(boxes_np):
+        cropped_img = crop_head_with_padding(image, box, padding_ratio=0.1)
+        filename = f"face_{i}.jpg"
+        save_path = os.path.join(output_dir, filename)
+        cropped_img.save(save_path)
+        
+        task = loop.run_in_executor(
+            None, 
+            analyze_demographics_with_qwen, 
+            save_path, 
+            qwen_model, 
+            analysis_prompt
+        )
+        tasks.append(task)
+        
+        results.append({
+            "filename": filename,
+            "relative_path": f"results/{request_id}/{filename}",
+            "score": float(scores_list[i]) if i < len(scores_list) else 0.0
+        })
+        
+    if tasks:
+        analysis_results = await asyncio.gather(*tasks)
+    else:
+        analysis_results = []
+        
+    for i, item in enumerate(results):
+        item["analysis"] = analysis_results[i]
+        
+    return {
+        "status": "success",
+        "message": f"成功检测并分析 {detected_count} 个人脸",
+        "detected_count": detected_count,
+        "request_id": request_id,
+        "full_visualization": full_vis_relative_path, 
+        "scores": scores_list, 
+        "results": results
+    }
+
+# 保留旧的同步接口以兼容其他潜在调用者，但内部实现可能会有问题如果它在 loop 中运行
+# 既然我们主要关注 fastAPI_tarot.py，我们可以直接替换 process_face_segmentation_and_analysis
+# 或者让它只是一个 wrapper
+def process_face_segmentation_and_analysis(
+    processor, 
+    image: Image.Image, 
+    prompt: str = "head", 
+    output_base_dir: str = "static/results",
+    qwen_model: str = "qwen-vl-max",
+    analysis_prompt: str = None
+) -> dict:
+    """
+    同步版本 (保留以兼容)
+    注意：如果在 async loop 中调用此函数，且此函数内部没有异步操作，则会阻塞 loop。
+    如果需要异步并发，请使用 process_face_segmentation_and_analysis_async
+    """
+    # 这里我们简单地复用逻辑，但去除异步部分，退化为串行
+    
+    # 1. SAM3 推理
+    inference_state = processor.set_image(image)
+    output = processor.set_text_prompt(state=inference_state, prompt=prompt)
+    masks, boxes, scores = output["masks"], output["boxes"], output["scores"]
+    
+    detected_count = len(masks)
+    if detected_count == 0:
+        return {
+            "status": "success",
+            "message": "未检测到目标",
+            "detected_count": 0,
+            "results": []
+        }
+        
+    request_id = f"{int(time.time())}_{uuid.uuid4().hex[:8]}"
+    output_dir = os.path.join(output_base_dir, request_id)
+    os.makedirs(output_dir, exist_ok=True)
+    
+    vis_filename = f"seg_{uuid.uuid4().hex}.jpg"
+    vis_path = os.path.join(output_dir, vis_filename)
+    try:
+        create_highlighted_visualization(image, masks, vis_path)
+        full_vis_relative_path = f"results/{request_id}/{vis_filename}"
+    except Exception as e:
+        print(f"可视化生成失败: {e}")
+        full_vis_relative_path = None
+
+    if isinstance(boxes, torch.Tensor):
+        boxes_np = boxes.cpu().numpy()
+    else:
+        boxes_np = boxes
+    
+    if isinstance(scores, torch.Tensor):
+        scores_list = scores.tolist()
+    else:
+        scores_list = scores if isinstance(scores, list) else [float(scores)]
+
+    results = []
+    for i, box in enumerate(boxes_np):
+        cropped_img = crop_head_with_padding(image, box, padding_ratio=0.1)
+        filename = f"face_{i}.jpg"
+        save_path = os.path.join(output_dir, filename)
+        cropped_img.save(save_path)
+        
+        # 同步调用
         analysis = analyze_demographics_with_qwen(save_path, model_name=qwen_model, prompt_template=analysis_prompt)
         
-        # 构造返回结果
-        # 注意：URL 生成需要依赖外部的 request context，这里只返回相对路径或文件名
-        # 由调用方组装完整 URL
         results.append({
             "filename": filename,
             "relative_path": f"results/{request_id}/{filename}",
@@ -232,7 +465,8 @@ def process_face_segmentation_and_analysis(
         "message": f"成功检测并分析 {detected_count} 个人脸",
         "detected_count": detected_count,
         "request_id": request_id,
-        "full_visualization": full_vis_relative_path, # 返回相对路径
+        "full_visualization": full_vis_relative_path, 
-        "scores": scores_list, # 返回全部分数
+        "scores": scores_list, 
         "results": results
     } 
+

static/admin.html

@@ -144,7 +144,7 @@
                         </div>
                         <div>
                             <h2 class="font-bold text-lg leading-tight tracking-tight">SAM3 Admin</h2>
-                            <p class="text-[10px] text-slate-400 font-bold tracking-widest uppercase mt-0.5">Quant Speed AI</p>
+                            <p class="text-[10px] text-slate-400 font-bold tracking-widest uppercase mt-0.5">Quantum Track AI</p>
                         </div>
                     </div>
                     
@@ -155,6 +155,11 @@
                             <i class="fas fa-chart-pie w-5 text-center transition-colors group-hover:text-blue-600" :class="currentTab === 'dashboard' ? 'text-blue-600' : 'text-slate-400'"></i>
                             <span class="font-medium">数据看板</span>
                         </a>
+                        <a href="#" @click.prevent="switchTab('tarot')" :class="{ 'active': currentTab === 'tarot' }" 
+                           class="nav-link flex items-center gap-3 px-4 py-3 text-slate-600 hover:bg-slate-50 hover:text-blue-600 rounded-xl transition-all duration-200 group">
+                            <i class="fas fa-star w-5 text-center transition-colors group-hover:text-blue-600" :class="currentTab === 'tarot' ? 'text-blue-600' : 'text-slate-400'"></i>
+                            <span class="font-medium">塔罗牌识别</span>
+                        </a>
                         <a href="#" @click.prevent="switchTab('gpu')" :class="{ 'active': currentTab === 'gpu' }" 
                            class="nav-link flex items-center gap-3 px-4 py-3 text-slate-600 hover:bg-slate-50 hover:text-blue-600 rounded-xl transition-all duration-200 group">
                             <i class="fas fa-microchip w-5 text-center transition-colors group-hover:text-blue-600" :class="currentTab === 'gpu' ? 'text-blue-600' : 'text-slate-400'"></i>
@@ -372,6 +377,158 @@
 
                         </div>
 
+                        <!-- Tarot Tab -->
+                        <div v-if="currentTab === 'tarot'" key="tarot" class="space-y-6">
+                            <!-- Input Section -->
+                            <div class="bg-white rounded-2xl shadow-sm border border-slate-100 p-6">
+                                <h3 class="font-bold text-slate-800 mb-6 flex items-center gap-2">
+                                    <span class="w-1 h-5 bg-purple-500 rounded-full"></span>
+                                    塔罗牌识别任务
+                                </h3>
+                                
+                                <div class="grid grid-cols-1 md:grid-cols-2 gap-6">
+                                    <!-- Image Upload -->
+                                    <div class="space-y-4">
+                                        <label class="block text-sm font-medium text-slate-700">上传图片</label>
+                                        <div class="flex items-center justify-center w-full">
+                                            <label for="dropzone-file" class="flex flex-col items-center justify-center w-full h-64 border-2 border-slate-300 border-dashed rounded-xl cursor-pointer bg-slate-50 hover:bg-slate-100 transition-colors relative overflow-hidden">
+                                                <div v-if="!tarotFile && !tarotImageUrl" class="flex flex-col items-center justify-center pt-5 pb-6">
+                                                    <i class="fas fa-cloud-upload-alt text-4xl text-slate-400 mb-3"></i>
+                                                    <p class="mb-2 text-sm text-slate-500"><span class="font-semibold">点击上传</span> 或拖拽文件到此处</p>
+                                                    <p class="text-xs text-slate-400">支持 JPG, PNG (MAX. 10MB)</p>
+                                                </div>
+                                                <div v-else class="absolute inset-0 flex items-center justify-center bg-slate-100">
+                                                    <img v-if="tarotPreview" :src="tarotPreview" class="max-h-full max-w-full object-contain">
+                                                    <div v-else class="text-slate-500 flex flex-col items-center">
+                                                        <i class="fas fa-link text-2xl mb-2"></i>
+                                                        <span class="text-xs truncate max-w-[200px]">{{ tarotImageUrl }}</span>
+                                                    </div>
+                                                    <button @click.prevent="clearTarotInput" class="absolute top-2 right-2 bg-white/80 p-1.5 rounded-full hover:bg-white text-slate-600 shadow-sm">
+                                                        <i class="fas fa-times"></i>
+                                                    </button>
+                                                </div>
+                                                <input id="dropzone-file" type="file" class="hidden" accept="image/*" @change="handleTarotFileChange" />
+                                            </label>
+                                        </div>
+                                        
+                                        <div class="relative">
+                                            <div class="absolute inset-y-0 left-0 flex items-center pl-3 pointer-events-none">
+                                                <i class="fas fa-link text-slate-400"></i>
+                                            </div>
+                                            <input type="text" v-model="tarotImageUrl" @input="handleUrlInput" 
+                                                class="bg-slate-50 border border-slate-200 text-slate-900 text-sm rounded-xl focus:ring-purple-500 focus:border-purple-500 block w-full pl-10 p-2.5 outline-none" 
+                                                placeholder="或者输入图片 URL...">
+                                        </div>
+                                    </div>
+
+                                    <!-- Settings -->
+                                    <div class="space-y-6">
+                                        <div>
+                                            <label class="block text-sm font-medium text-slate-700 mb-2">预期卡牌数量</label>
+                                            <div class="flex items-center gap-4">
+                                                <input type="number" v-model.number="tarotExpectedCount" min="1" max="10" 
+                                                    class="bg-slate-50 border border-slate-200 text-slate-900 text-sm rounded-xl focus:ring-purple-500 focus:border-purple-500 block w-full p-2.5 outline-none font-mono">
+                                                <span class="text-sm text-slate-500 whitespace-nowrap">张</span>
+                                            </div>
+                                            <p class="text-xs text-slate-400 mt-2">系统将尝试检测并分割指定数量的卡牌。如果检测数量不符，将返回错误提示。</p>
+                                        </div>
+
+                                        <div class="pt-4">
+                                            <button @click="recognizeTarot" :disabled="isRecognizing || (!tarotFile && !tarotImageUrl)" 
+                                                class="w-full bg-gradient-to-r from-purple-600 to-indigo-600 hover:from-purple-700 hover:to-indigo-700 text-white font-bold py-3 px-4 rounded-xl shadow-lg shadow-purple-500/30 transition-all transform active:scale-[0.98] disabled:opacity-50 disabled:cursor-not-allowed flex items-center justify-center gap-2">
+                                                <i class="fas fa-magic" :class="{'fa-spin': isRecognizing}"></i>
+                                                {{ isRecognizing ? '正在识别中...' : '开始识别 (Recognize)' }}
+                                            </button>
+                                        </div>
+                                    </div>
+                                </div>
+                            </div>
+
+                            <!-- Results Section -->
+                            <div v-if="tarotResult" class="space-y-6 animate-fade-in">
+                                <!-- Status Banner -->
+                                <div :class="tarotResult.status === 'success' ? 'bg-green-50 border-green-200 text-green-700' : 'bg-red-50 border-red-200 text-red-700'" 
+                                    class="p-4 rounded-xl border flex items-center gap-3 shadow-sm">
+                                    <i :class="tarotResult.status === 'success' ? 'fas fa-check-circle' : 'fas fa-exclamation-circle'" class="text-xl"></i>
+                                    <div>
+                                        <h4 class="font-bold">{{ tarotResult.status === 'success' ? '识别成功' : '识别失败' }}</h4>
+                                        <p class="text-sm opacity-90">{{ tarotResult.message }}</p>
+                                    </div>
+                                </div>
+
+                                <!-- Spread Info -->
+                                <div v-if="tarotResult.spread_info" class="bg-white rounded-2xl shadow-sm border border-slate-100 p-6">
+                                    <h3 class="font-bold text-slate-800 mb-4 flex items-center gap-2">
+                                        <i class="fas fa-layer-group text-purple-500"></i>
+                                        牌阵信息
+                                    </h3>
+                                    <div class="bg-purple-50 rounded-xl p-4 border border-purple-100">
+                                        <div class="flex flex-col md:flex-row gap-4">
+                                            <div class="md:w-1/3">
+                                                <span class="text-xs font-bold text-purple-400 uppercase tracking-wider">牌阵名称</span>
+                                                <div class="text-xl font-bold text-slate-800 mt-1">{{ tarotResult.spread_info.spread_name }}</div>
+                                            </div>
+                                            <div class="md:w-2/3 border-t md:border-t-0 md:border-l border-purple-200 pt-4 md:pt-0 md:pl-4">
+                                                <span class="text-xs font-bold text-purple-400 uppercase tracking-wider">描述 / 寓意</span>
+                                                <div class="text-sm text-slate-700 mt-1 leading-relaxed">{{ tarotResult.spread_info.description || '暂无描述' }}</div>
+                                            </div>
+                                        </div>
+                                         <div v-if="tarotResult.spread_info.model_used" class="mt-3 pt-3 border-t border-purple-200 flex items-center gap-2">
+                                            <span class="text-xs bg-white text-purple-600 px-2 py-0.5 rounded border border-purple-200 font-mono">
+                                                <i class="fas fa-robot mr-1"></i>{{ tarotResult.spread_info.model_used }}
+                                            </span>
+                                        </div>
+                                    </div>
+                                </div>
+
+                                <!-- Cards Grid -->
+                                <div v-if="tarotResult.tarot_cards && tarotResult.tarot_cards.length > 0" class="grid grid-cols-1 md:grid-cols-2 lg:grid-cols-3 gap-6">
+                                    <div v-for="(card, index) in tarotResult.tarot_cards" :key="index" class="bg-white rounded-2xl shadow-sm border border-slate-100 overflow-hidden hover:shadow-md transition-shadow group">
+                                        <div class="relative aspect-[2/3] bg-slate-100 overflow-hidden cursor-pointer" @click="previewImage(card.url)">
+                                            <img :src="card.url" class="w-full h-full object-cover group-hover:scale-105 transition-transform duration-500">
+                                            <div class="absolute top-2 right-2 bg-black/60 backdrop-blur-sm text-white text-xs font-bold px-2 py-1 rounded">
+                                                #{{ index + 1 }}
+                                            </div>
+                                        </div>
+                                        <div class="p-4 space-y-3">
+                                            <div class="flex justify-between items-start">
+                                                <div>
+                                                    <h4 class="font-bold text-lg text-slate-800">{{ card.recognition?.name || '未知' }}</h4>
+                                                    <div class="flex items-center gap-2 mt-1">
+                                                        <span :class="card.recognition?.position === '正位' ? 'bg-green-100 text-green-700' : 'bg-red-100 text-red-700'" 
+                                                            class="text-xs font-bold px-2 py-0.5 rounded">
+                                                            {{ card.recognition?.position || '未知' }}
+                                                        </span>
+                                                        <span v-if="card.is_rotated" class="text-[10px] text-slate-400 bg-slate-100 px-1.5 rounded" title="已自动矫正方向">
+                                                            <i class="fas fa-sync-alt"></i> Auto-Rotated
+                                                        </span>
+                                                    </div>
+                                                </div>
+                                            </div>
+                                            
+                                            <div v-if="card.recognition?.model_used" class="pt-3 border-t border-slate-100 flex items-center justify-between text-xs">
+                                                <span class="text-slate-400">Model Used:</span>
+                                                <span class="font-mono text-purple-600 bg-purple-50 px-1.5 py-0.5 rounded border border-purple-100">
+                                                    {{ card.recognition.model_used }}
+                                                </span>
+                                            </div>
+                                        </div>
+                                    </div>
+                                </div>
+
+                                <!-- Full Visualization -->
+                                <div v-if="tarotResult.full_visualization" class="bg-white rounded-2xl shadow-sm border border-slate-100 p-6">
+                                    <h3 class="font-bold text-slate-800 mb-4 flex items-center gap-2">
+                                        <i class="fas fa-image text-blue-500"></i>
+                                        整体可视化结果
+                                    </h3>
+                                    <div class="rounded-xl overflow-hidden border border-slate-200 cursor-pointer" @click="previewImage(tarotResult.full_visualization)">
+                                        <img :src="tarotResult.full_visualization" class="w-full h-auto">
+                                    </div>
+                                </div>
+                            </div>
+                        </div>
+
                         <!-- History Tab -->
                         <div v-if="currentTab === 'history'" key="history" class="bg-white rounded-2xl shadow-sm border border-slate-100 overflow-hidden">
                              <div class="overflow-x-auto">
@@ -383,7 +540,7 @@
                                             <th class="px-6 py-4">Prompt / 详情</th>
                                             <th class="px-6 py-4 text-center">耗时</th>
                                             <th class="px-6 py-4 text-center">状态</th>
-                                            <th class="px-6 py-4 text-center">查看</th>
+                                            <th class="px-6 py-4 text-center">操作</th>
                                         </tr>
                                     </thead>
                                     <tbody class="divide-y divide-slate-100">
@@ -788,6 +945,14 @@
                 const gpuHistory = ref([]);
                 let gpuInterval = null;
 
+                // Tarot State
+                const tarotFile = ref(null);
+                const tarotImageUrl = ref('');
+                const tarotExpectedCount = ref(3);
+                const tarotPreview = ref(null);
+                const tarotResult = ref(null);
+                const isRecognizing = ref(false);
+                
                 // Filters
                 const selectedTimeRange = ref('all');
                 const selectedType = ref('all');
@@ -1147,6 +1312,90 @@
                     } catch (e) { alert('删除失败: ' + e.message); }
                 };
 
+                // --- Tarot Actions ---
+                const handleTarotFileChange = (event) => {
+                    const file = event.target.files[0];
+                    if (file) {
+                        tarotFile.value = file;
+                        tarotImageUrl.value = ''; // Clear URL if file is selected
+                        tarotPreview.value = URL.createObjectURL(file);
+                        tarotResult.value = null; // Clear previous result
+                    }
+                };
+
+                const handleUrlInput = () => {
+                    if (tarotImageUrl.value) {
+                        tarotFile.value = null; // Clear file if URL is entered
+                        tarotPreview.value = null; // Can't preview external URL easily without loading it, or just use the URL
+                        // Simple preview for URL
+                        tarotPreview.value = tarotImageUrl.value;
+                        tarotResult.value = null;
+                    } else {
+                        tarotPreview.value = null;
+                    }
+                };
+
+                const clearTarotInput = () => {
+                    tarotFile.value = null;
+                    tarotImageUrl.value = '';
+                    tarotPreview.value = null;
+                    tarotResult.value = null;
+                    // Reset file input value
+                    const fileInput = document.getElementById('dropzone-file');
+                    if (fileInput) fileInput.value = '';
+                };
+
+                const recognizeTarot = async () => {
+                    if (!tarotFile.value && !tarotImageUrl.value) return;
+                    
+                    isRecognizing.value = true;
+                    tarotResult.value = null;
+                    
+                    try {
+                        const formData = new FormData();
+                        if (tarotFile.value) {
+                            formData.append('file', tarotFile.value);
+                        } else {
+                            formData.append('image_url', tarotImageUrl.value);
+                        }
+                        formData.append('expected_count', tarotExpectedCount.value);
+                        
+                        // Use axios directly or a helper. Need to handle API Key if required by backend, 
+                        // but admin usually has session. Wait, the backend endpoints like /recognize_tarot 
+                        // require X-API-Key header.
+                        // The admin page uses cookie for /admin/api/* but /recognize_tarot is a public API protected by Key.
+                        // We should add the key to the header.
+                        
+                        const config = {
+                            headers: {
+                                'X-API-Key': '123quant-speed' // Hardcoded as per fastAPI_tarot.py VALID_API_KEY
+                            },
+                            timeout: 120000 // 2分钟超时，大模型响应较慢
+                        };
+                        
+                        const res = await axios.post('/recognize_tarot', formData, config);
+                        tarotResult.value = res.data;
+                        
+                    } catch (e) {
+                        console.error(e);
+                        let msg = e.response?.data?.detail || e.message || '识别请求失败';
+                        
+                        // 针对 504 Gateway Timeout 或 请求超时做特殊提示
+                        if (e.response && e.response.status === 504) {
+                            msg = '请求超时 (504)：大模型处理时间较长。后台可能仍在运行，请稍后在“识别记录”中刷新查看结果。';
+                        } else if (e.code === 'ECONNABORTED') {
+                            msg = '请求超时：网络连接中断或服务器响应过慢。请稍后重试。';
+                        }
+
+                        tarotResult.value = {
+                            status: 'failed',
+                            message: msg
+                        };
+                    } finally {
+                        isRecognizing.value = false;
+                    }
+                };
+
                 // --- Navigation & Helpers ---
                 const switchTab = (tab) => {
                     const prevTab = currentTab.value;
@@ -1236,6 +1485,7 @@
                 const getPageTitle = (tab) => {
                     const map = {
                         'dashboard': '数据看板',
+                        'tarot': '塔罗牌识别',
                         'history': '识别记录',
                         'files': '文件资源管理',
                         'prompts': '提示词工程',
@@ -1248,6 +1498,7 @@
                 const getPageSubtitle = (tab) => {
                     const map = {
                         'dashboard': '系统运行状态与核心指标概览',
+                        'tarot': 'SAM3 + Qwen-VL 联合识别与分割',
                         'history': '所有视觉识别任务的历史流水',
                         'files': '查看和管理生成的图像及JSON结果',
                         'prompts': '调整各个识别场景的 System Prompt',
@@ -1460,7 +1711,10 @@
                     selectedTimeRange, selectedType,
                     barChartRef, pieChartRef, promptPieChartRef, promptBarChartRef, wordCloudRef,
                     formatBytes, gpuStatus,
-                    gpuUtilChartRef, gpuTempChartRef
+                    gpuUtilChartRef, gpuTempChartRef,
+                    // Tarot
+                    tarotFile, tarotImageUrl, tarotExpectedCount, tarotPreview, tarotResult, isRecognizing,
+                    handleTarotFileChange, handleUrlInput, clearTarotInput, recognizeTarot
                 };
             }
         }).mount('#app');