Update app.py

app.py

@@ -5,6 +5,7 @@ import uuid
 import csv
 from datetime import datetime
 from pathlib import Path
+from PIL import Image  # 引入 PIL 用于处理图片
 from huggingface_hub import CommitScheduler, snapshot_download
 
 # --- 1. 配置区域 ---
@@ -14,7 +15,7 @@ LOG_FOLDER = Path("logs")
 LOG_FOLDER.mkdir(parents=True, exist_ok=True)
 TOKEN = os.environ.get("HF_TOKEN")
 
-# --- 2. 自动下载数据 (保证不为空) ---
+# --- 2. 自动下载数据 ---
 if not os.path.exists(DATA_FOLDER) or not os.listdir(DATA_FOLDER):
     try:
         print("🚀 正在从 Dataset 下载数据...")
@@ -27,7 +28,7 @@ if not os.path.exists(DATA_FOLDER) or not os.listdir(DATA_FOLDER):
         )
         print("✅ 数据下载完成！")
     except Exception as e:
-        print(f"⚠️ 下载失败 (请检查 Token 或网络): {e}")
+        print(f"⚠️ 下载失败: {e}")
 
 # --- 3. 启动同步调度器 ---
 scheduler = CommitScheduler(
@@ -78,54 +79,75 @@ def load_data():
 
 ALL_GROUPS, ALL_GROUP_IDS = load_data()
 
+# --- NEW: 图片优化函数 (提速关键) ---
+def optimize_image(image_path, max_width=800):
+    """
+    读取图片并调整大小，减少传输时间。
+    max_width: 限制最大宽度为 800px (足够人眼评估)
+    """
+    if not image_path:
+        return None
+    try:
+        img = Image.open(image_path)
+        # 如果图片太大，就缩小
+        if img.width > max_width:
+            ratio = max_width / img.width
+            new_height = int(img.height * ratio)
+            img = img.resize((max_width, new_height), Image.LANCZOS)
+        return img
+    except Exception as e:
+        print(f"Error loading image {image_path}: {e}")
+        return None
+
 # --- 5. 核心逻辑 ---
 
 def get_next_question(user_state):
     """准备下一题的数据"""
     idx = user_state["index"]
     
-    # 1. 结束判断
     if idx >= len(ALL_GROUP_IDS):
         return (
-            gr.update(visible=False), # Origin
-            gr.update(visible=False), # Gallery
-            gr.update(visible=False), # Checkbox
-            gr.update(visible=False), # Instruction
-            gr.update(visible=False), # Submit Btn
-            gr.update(visible=False), # None Btn
+            gr.update(visible=False), 
+            gr.update(visible=False), 
+            gr.update(visible=False), 
+            gr.update(visible=False), 
+            gr.update(visible=False), 
+            gr.update(visible=False), 
             gr.update(value="## 🎉 测试结束！感谢您的参与。", visible=True),
             user_state,
-            [] # current_candidates_info
+            [] 
         )
     
-    # 2. 获取数据
     group_id = ALL_GROUP_IDS[idx]
     group_data = ALL_GROUPS[group_id]
     
-    # 3. 准备原图
-    origin_path = group_data["origin"]
+    # 1. 优化原图 (返回 PIL 对象而不是路径)
+    origin_img = optimize_image(group_data["origin"], max_width=600)
     
-    # 4. 准备候选图（打乱）
+    # 2. 优化候选图
     candidates = group_data["candidates"].copy()
     random.shuffle(candidates)
     
-    # 构造 Gallery 数据 [(path, label), ...] 和 选项列表 ["Option A", ...]
     gallery_items = []
     choices = []
-    candidates_info = [] # 用于存储真实路径，方便后续查找
+    candidates_info = [] 
     
     for i, path in enumerate(candidates):
-        label = f"Option {chr(65+i)}" # Option A, Option B...
-        gallery_items.append((path, label))
+        label = f"Option {chr(65+i)}"
+        
+        # 优化每张候选图
+        optimized_img = optimize_image(path, max_width=600)
+        
+        gallery_items.append((optimized_img, label))
         choices.append(label)
         candidates_info.append({"label": label, "path": path})
         
     instruction = f"### 任务 ({idx + 1} / {len(ALL_GROUP_IDS)})\n\n{group_data['instruction']}"
     
     return (
-        gr.update(value=origin_path, visible=True if origin_path else False),
+        gr.update(value=origin_img, visible=True if origin_img else False),
         gr.update(value=gallery_items, visible=True),
-        gr.update(choices=choices, value=[], visible=True), # 重置多选框
+        gr.update(choices=choices, value=[], visible=True), 
         gr.update(value=instruction, visible=True),
         gr.update(visible=True),
         gr.update(visible=True),
@@ -135,38 +157,30 @@ def get_next_question(user_state):
     )
 
 def save_and_next(user_state, candidates_info, selected_options, is_none=False):
-    """保存并进入下一题"""
     current_idx = user_state["index"]
     group_id = ALL_GROUP_IDS[current_idx]
     
-    # --- 保存逻辑 ---
     if is_none:
         choice_str = "Rejected All"
         method_str = "None_Satisfied"
     else:
-        # 检查是否未选
         if not selected_options:
             raise gr.Error("请至少勾选一个选项，或点击“都不满意”")
             
         choice_str = "; ".join(selected_options)
-        
-        # 查找对应的方法名
         selected_methods = []
-        for opt in selected_options: # opt is "Option A"
-            # 找到对应的文件路径
+        for opt in selected_options: 
             for info in candidates_info:
                 if info["label"] == opt:
                     path = info["path"]
                     filename = os.path.basename(path)
                     name = os.path.splitext(filename)[0]
-                    # 简单清洗文件名拿到方法名
                     parts = name.split('_', 1)
                     method = parts[1] if len(parts) > 1 else name
                     selected_methods.append(method)
                     break
         method_str = "; ".join(selected_methods)
 
-    # 写入 CSV
     user_file = LOG_FOLDER / f"user_{user_state['user_id']}.csv"
     with scheduler.lock:
         exists = user_file.exists()
@@ -182,42 +196,35 @@ def save_and_next(user_state, candidates_info, selected_options, is_none=False):
                 method_str
             ])
             
-    # --- 状态更新 ---
     user_state["index"] += 1
-    
-    # --- 加载下一题 ---
     return get_next_question(user_state)
 
 # --- 6. 界面构建 ---
 with gr.Blocks(title="User Study") as demo:
     
-    # 状态变量
-    state_user = gr.State({"user_id": str(uuid.uuid4())[:8], "index": 0})
-    state_candidates_info = gr.State([]) # 存当前页面的候选图信息，用于把 Option A 映射回文件名
+    state_user = gr.State(lambda: {"user_id": str(uuid.uuid4())[:8], "index": 0})
+    state_candidates_info = gr.State([]) 
     
-    # 布局
     with gr.Row():
         md_instruction = gr.Markdown("Loading...")
     
     with gr.Row():
-        # 左侧：原图
         with gr.Column(scale=1):
-            img_origin = gr.Image(label="Reference (参考原图)", interactive=False, height=400)
+            # 将 format 设置为 jpeg 进一步减小体积
+            img_origin = gr.Image(label="Reference (参考原图)", interactive=False, height=400, format="jpeg")
         
-        # 右侧：候选图 + 选择区
         with gr.Column(scale=2):
-            # 1. Gallery 显示所有选项
             gallery_candidates = gr.Gallery(
                 label="Candidates (候选结果)", 
                 columns=[2], 
                 height="auto",
                 object_fit="contain",
-                interactive=False # 禁止 Gallery 自身的点击选择，用下面的 Checkbox 替代
+                interactive=False,
+                format="jpeg" # 强制输出 JPEG 格式
             )
             
             gr.Markdown("👇 **请在下方勾选您认为最好的结果（可多选）：**")
             
-            # 2. 多选框 (核心交互)
             checkbox_options = gr.CheckboxGroup(
                 choices=[], 
                 label="您的选择", 
@@ -230,23 +237,18 @@ with gr.Blocks(title="User Study") as demo:
     
     md_end = gr.Markdown(visible=False)
 
-    # --- 事件绑定 ---
-    
-    # 1. 页面加载时，加载第一题
     demo.load(
         fn=get_next_question,
         inputs=[state_user],
         outputs=[img_origin, gallery_candidates, checkbox_options, md_instruction, btn_submit, btn_none, md_end, state_user, state_candidates_info]
     )
     
-    # 2. 提交按钮
     btn_submit.click(
         fn=lambda s, c, o: save_and_next(s, c, o, is_none=False),
         inputs=[state_user, state_candidates_info, checkbox_options],
         outputs=[img_origin, gallery_candidates, checkbox_options, md_instruction, btn_submit, btn_none, md_end, state_user, state_candidates_info]
     )
     
-    # 3. 都不满意按钮
     btn_none.click(
         fn=lambda s, c, o: save_and_next(s, c, o, is_none=True),
         inputs=[state_user, state_candidates_info, checkbox_options],