fix: remove MBench-V tabs, fix white-on-white text in aux info

app.py

@@ -228,8 +228,12 @@ def _render_mbench_a_aux(task: dict) -> str:
     """Render auxiliary HTML info based on task subset."""
     subset = task["subset"]
 
+    # Shared style: visible on both light and dark themes
+    box = 'style="background:#e8edf2;color:#1a1a1a;padding:12px;border-radius:8px;margin-bottom:12px;border:1px solid #c0c8d0"'
+    label_style = 'style="margin:0 0 4px;color:#1a1a1a"'
+    text_style = 'style="margin:0;color:#2a2a2a;font-size:14px;line-height:1.5"'
+
     if subset == "object":
-        # Show mask overlay image from HF
         sample_id = task["sample_id"]
         gt_cat = MBENCH_A_GT_CATEGORY_MAP[subset]
         first_frame_url = hf_hub_url(
@@ -238,9 +242,9 @@ def _render_mbench_a_aux(task: dict) -> str:
             repo_type="dataset",
         )
         return (
-            f'<div style="background:#f8f9fa;padding:12px;border-radius:8px;margin-bottom:12px">'
-            f'<p style="margin:0 0 8px"><b>🎯 请关注画面中被标注（高亮）的物体</b></p>'
-            f'<img src="{first_frame_url}" style="max-width:100%;max-height:250px;border-radius:4px">'
+            f'<div {box}>'
+            f'<p {label_style}><b>🎯 请关注画面中被标注（高亮）的物体</b></p>'
+            f'<img src="{first_frame_url}" style="max-width:100%;max-height:250px;border-radius:4px;margin-top:8px">'
             f'</div>'
         )
 
@@ -248,35 +252,34 @@ def _render_mbench_a_aux(task: dict) -> str:
         caption = task.get("caption", "(无描述)")
         motion = task.get("camera_motion", "left_then_right")
         motion_desc = task.get("camera_motion_description", motion)
-        # Camera GIF - use HF asset URL
         gif_url = _mbench_a_asset_hf_url(f"camera_diagrams/{motion}.gif")
         return (
-            f'<div style="background:#f8f9fa;padding:12px;border-radius:8px;margin-bottom:12px">'
+            f'<div {box}>'
             f'<div style="display:flex;gap:16px;flex-wrap:wrap">'
             f'<div style="flex:0 0 200px">'
-            f'<p style="margin:0 0 4px"><b>🎬 预期相机运动</b></p>'
-            f'<p style="margin:0 0 8px;color:#555">{motion_desc}</p>'
-            f'<img src="{gif_url}" style="width:180px;border-radius:4px;border:1px solid #ddd">'
+            f'<p {label_style}><b>🎬 预期相机运动</b></p>'
+            f'<p style="margin:0 0 8px;color:#2a2a2a">{motion_desc}</p>'
+            f'<img src="{gif_url}" style="width:180px;border-radius:4px;border:1px solid #aaa">'
             f'</div>'
             f'<div style="flex:1;min-width:250px">'
-            f'<p style="margin:0 0 4px"><b>📝 场景描述</b></p>'
-            f'<p style="margin:0;color:#333;font-size:14px;line-height:1.5">{caption}</p>'
+            f'<p {label_style}><b>📝 场景描述</b></p>'
+            f'<p {text_style}>{caption}</p>'
             f'</div>'
             f'</div></div>'
         )
 
     elif subset == "human":
         return (
-            '<div style="background:#f8f9fa;padding:12px;border-radius:8px;margin-bottom:12px">'
-            '<p style="margin:0"><b>👤 请关注视频中的人物</b>：观察人物离开画面再回来后，面部和外观是否保持一致。</p>'
-            '</div>'
+            f'<div {box}>'
+            f'<p {text_style}><b>👤 请关注视频中的人物</b>：观察人物离开画面再回来后，面部和外观是否保持一致。</p>'
+            f'</div>'
         )
 
     else:  # environment
         return (
-            '<div style="background:#f8f9fa;padding:12px;border-radius:8px;margin-bottom:12px">'
-            '<p style="margin:0"><b>🏞️ 请关注整体场景</b>：观察相机转回来后，场景的布局、风格、光照是否保持一致。</p>'
-            '</div>'
+            f'<div {box}>'
+            f'<p {text_style}><b>🏞️ 请关注整体场景</b>：观察相机转回来后，场景的布局、风格、光照是否保持一致。</p>'
+            f'</div>'
         )
 
 # ---------------------------------------------------------------------------
@@ -760,77 +763,7 @@ with gr.Blocks(title="MBench 标注", theme=gr.themes.Soft(), css=CUSTOM_CSS) as
     gr.Markdown("# 🎬 MBench 视频标注平台")
 
     with gr.Tabs():
-        # ═══════════════ Tab 1: 单视频二值标注 (MBench-V) ═══════════════
-        with gr.Tab("单视频标注 (Memory Issue)"):
-            gr.Markdown("判断该视频是否出现了**记忆问题**（人物/物体离开后回来变了样、场景前后不一致等）。")
-            b_stats = gr.Markdown("")
-            b_state = gr.State({})
-            with gr.Row():
-                b_name = gr.Textbox(label="标注员名字", placeholder="例如: alice", scale=4)
-                b_login = gr.Button("开始", variant="primary", scale=1)
-            b_status = gr.Markdown("")
-            with gr.Row():
-                with gr.Column(scale=3):
-                    b_video = gr.HTML("<p>请先登录。</p>")
-                with gr.Column(scale=2):
-                    b_prompt = gr.Textbox(label="Prompt（5 段）", lines=14, interactive=False, elem_id="prompt_box")
-            b_meta = gr.Markdown("")
-            with gr.Row():
-                b_verdict = gr.Radio(["否", "是"], value="否", label="是否出现记忆问题？")
-                b_note = gr.Textbox(label="备注（可选）", lines=1)
-            with gr.Row():
-                b_submit = gr.Button("✅ 提交并下一条", variant="primary")
-                b_skip = gr.Button("⏭️ 跳过")
-            b_login_outs = [b_state, b_video, b_meta, b_prompt, b_status, b_stats]
-            b_step_outs = [b_state, b_video, b_meta, b_prompt, b_verdict, b_note, b_status, b_stats]
-            b_login.click(binary_start, [b_name, b_state], b_login_outs)
-            b_name.submit(binary_start, [b_name, b_state], b_login_outs)
-            b_submit.click(binary_submit, [b_state, b_verdict, b_note], b_step_outs)
-            b_skip.click(binary_skip, [b_state], b_step_outs)
-
-        # ═══════════════ Tab 2: MBench-V Pairwise ═══════════════
-        with gr.Tab("MBench-V Pairwise"):
-            gr.Markdown(
-                "同一 Prompt 下**两个模型的视频并排**，选择哪个在指定维度上更好。\n\n"
-                "视频 A/B 的模型身份已匿名随机分配。"
-            )
-            p_stats = gr.Markdown("")
-            p_state = gr.State({})
-            with gr.Row():
-                p_name = gr.Textbox(label="标注员名字", placeholder="例如: bob", scale=3)
-                p_dim = gr.Dropdown(
-                    choices=[(dl, dk) for dk, dl, _ in PAIRWISE_DIMENSIONS],
-                    value=PAIRWISE_DIMENSIONS[0][0],
-                    label="评测维度", scale=2,
-                )
-                p_login = gr.Button("开始", variant="primary", scale=1)
-            p_status = gr.Markdown("")
-            with gr.Row(equal_height=True):
-                with gr.Column(scale=1, min_width=360):
-                    gr.Markdown("### 视频 A")
-                    p_video_a = gr.HTML("<p>请先登录。</p>")
-                with gr.Column(scale=1, min_width=360):
-                    gr.Markdown("### 视频 B")
-                    p_video_b = gr.HTML("<p>请先登录。</p>")
-            with gr.Row():
-                with gr.Column(scale=3):
-                    p_meta = gr.Markdown("")
-                with gr.Column(scale=2):
-                    p_prompt = gr.Textbox(label="Prompt", lines=10, interactive=False)
-            with gr.Row():
-                p_verdict = gr.Radio(["左边更好", "差不多", "右边更好"], value="差不多", label="哪个视频更好？")
-                p_note = gr.Textbox(label="备注（可选）", lines=1)
-            with gr.Row():
-                p_submit = gr.Button("✅ 提交并下一对", variant="primary")
-                p_skip = gr.Button("⏭️ 跳过")
-            p_login_outs = [p_state, p_video_a, p_video_b, p_meta, p_prompt, p_status, p_stats]
-            p_step_outs = [p_state, p_video_a, p_video_b, p_meta, p_prompt, p_status, p_stats]
-            p_login.click(pairwise_start, [p_name, p_dim, p_state], p_login_outs)
-            p_name.submit(pairwise_start, [p_name, p_dim, p_state], p_login_outs)
-            p_submit.click(pairwise_submit, [p_state, p_verdict, p_note], p_step_outs)
-            p_skip.click(pairwise_skip, [p_state], p_step_outs)
-
-        # ═══════════════ Tab 3: MBench-A Pairwise ═══════════════
+        # ═══════════════ MBench-A Pairwise ═══════════════
         with gr.Tab("MBench-A 对比 (World Models)"):
             gr.Markdown(
                 "## 🌍 MBench-A — 世界模型记忆能力评测\n\n"