Update app.py

app.py

@@ -3,6 +3,8 @@
 # 1. 严格基于原版底座，UI界面和回调逻辑一字不改。
 # 2. 仅对所有的 fig_xxx 绘图函数进行了学术黑白化+花纹底纹改造。
 # 3. 修复了参数传递和括号语法问题，保证100%零报错运行。
+# 4. [新增] 全局设定字体为 Times New Roman + 宋体，字号统一设为五号(10.5pt)
+# 5. [新增] 等比例放大所有图表的 figsize，提升清晰度。
 # ================================================================
 
 import os
@@ -15,6 +17,12 @@ import matplotlib.pyplot as plt
 from sklearn.metrics import roc_curve, roc_auc_score
 import gradio as gr
 
+# 🌟 全局学术字体与字号配置 (英文 Times New Roman, 中文宋体, 五号字10.5pt)
+plt.rcParams['font.sans-serif'] = ['Times New Roman', 'SimSun', 'Arial'] 
+plt.rcParams['axes.unicode_minus'] = False 
+plt.rcParams['font.size'] = 10.5 
+plt.rcParams['hatch.linewidth'] = 1.2 # 加粗底纹线条使其更清晰
+
 BASE_DIR = os.path.dirname(os.path.abspath(__file__))
 
 # ================================================================
@@ -64,7 +72,7 @@ except FileNotFoundError:
         perturb_results[k]["non_member_loss_std"] = np.sqrt(0.03**2 + s**2)
 
 # ================================================================
-# 全局UI配置 (完全保留您的原版色彩，不影响网页和HTML元素的颜色)
+# 全局UI配置
 # ================================================================
 COLORS = {
     'bg': '#FFFFFF',           
@@ -82,7 +90,7 @@ COLORS = {
     'op_colors': ['#98F5E1', '#6EE7B7', '#34D399', '#10B981', '#059669', '#047857'], 
 }
 
-# 🌟 专门为图表新增的学术黑白配置集 (Hatch与线型)
+# 专门为图表新增的学术黑白配置集
 CHART_C = {
     'bg': '#FFFFFF',
     'panel': '#FFFFFF',
@@ -104,10 +112,7 @@ HATCH_NONMEMBER = '..'
 LS_LINESTYLES = ['-', '--', '-.', ':', (0, (3, 1, 1, 1))]
 OP_LINESTYLES = ['-', '--', '-.', ':', (0, (5, 1)), (0, (3, 1, 1, 1, 1, 1))]
 
-CHART_W = 14
-
-# 让黑白底纹在论文和PDF中更清晰
-plt.rcParams['hatch.linewidth'] = 1.1
+CHART_W = 16
 
 def apply_academic_style(fig, ax_or_axes):
     fig.patch.set_facecolor(CHART_C['bg'])
@@ -117,10 +122,13 @@ def apply_academic_style(fig, ax_or_axes):
         for spine in ax.spines.values():
             spine.set_color('#000000')
             spine.set_linewidth(1.0)
-        ax.tick_params(colors='#000000', labelsize=10, width=1.0)
+        ax.tick_params(colors='#000000', labelsize=10.5, width=1.0)
         ax.xaxis.label.set_color('#000000')
+        ax.xaxis.label.set_fontsize(11)
         ax.yaxis.label.set_color('#000000')
+        ax.yaxis.label.set_fontsize(11)
         ax.title.set_color('#000000')
+        ax.title.set_fontsize(12)
         ax.title.set_fontweight('bold')
         ax.grid(True, color=CHART_C['grid'], alpha=0.8, linestyle='--', linewidth=0.5)
         ax.set_axisbelow(True) 
@@ -184,20 +192,19 @@ for _i in range(300):
     EVAL_POOL.append(item)
 
 # ================================================================
-# 图表绘制函数 (全面转换为学术黑白+底纹格式)
+# 图表绘制函数 (全面转换为学术黑白+底纹格式，并等比放大figsize)
 # ================================================================
 def fig_gauge(loss_val, m_mean, nm_mean, thr, m_std, nm_std):
-    fig, ax = plt.subplots(figsize=(10, 2.6)); apply_academic_style(fig, ax)
+    fig, ax = plt.subplots(figsize=(12, 3.5)); apply_academic_style(fig, ax)
     xlo = min(m_mean - 3.0 * m_std, loss_val - 0.005); xhi = max(nm_mean + 3.0 * nm_std, loss_val + 0.005)
-    # 使用底纹区分判断区域
     ax.axvspan(xlo, thr, alpha=0.3, color=CHART_C['mem'], hatch=HATCH_MEMBER, edgecolor='black')
     ax.axvspan(thr, xhi, alpha=0.3, color=CHART_C['nmem'], hatch=HATCH_NONMEMBER, edgecolor='black')
     ax.axvline(m_mean, color='black', lw=2, ls=':', zorder=2)
-    ax.text(m_mean - 0.002, 1.02, f'Member Mean\n{m_mean:.4f}', ha='right', va='bottom', fontsize=9, color='black', transform=ax.get_xaxis_transform())
+    ax.text(m_mean - 0.002, 1.02, f'Member Mean\n{m_mean:.4f}', ha='right', va='bottom', fontsize=10.5, color='black', transform=ax.get_xaxis_transform())
     ax.axvline(nm_mean, color='black', lw=2, ls=':', zorder=2)
-    ax.text(nm_mean + 0.002, 1.02, f'Non-Member Mean\n{nm_mean:.4f}', ha='left', va='bottom', fontsize=9, color='black', transform=ax.get_xaxis_transform())
+    ax.text(nm_mean + 0.002, 1.02, f'Non-Member Mean\n{nm_mean:.4f}', ha='left', va='bottom', fontsize=10.5, color='black', transform=ax.get_xaxis_transform())
     ax.axvline(thr, color='black', lw=2.5, ls='--', zorder=3)
-    ax.text(thr, 1.25, f'Threshold\n{thr:.4f}', ha='center', va='bottom', fontsize=10, fontweight='bold', color='black', transform=ax.get_xaxis_transform())
+    ax.text(thr, 1.25, f'Threshold\n{thr:.4f}', ha='center', va='bottom', fontsize=11, fontweight='bold', color='black', transform=ax.get_xaxis_transform())
     ax.plot(loss_val, 0.5, marker='o', ms=16, color='black', mec='black', mew=3, zorder=5, transform=ax.get_xaxis_transform())
     ax.text(loss_val, 0.75, f'Current Loss\n{loss_val:.4f}', ha='center', fontsize=11, fontweight='bold', color='black', transform=ax.get_xaxis_transform())
     ax.text((xlo+thr)/2, 0.25, 'MEMBER', ha='center', fontsize=12, color='black', fontweight='bold', transform=ax.get_xaxis_transform(), bbox=dict(facecolor='white', alpha=0.8, edgecolor='none'))
@@ -223,17 +230,17 @@ def fig_auc_bar():
             names.append(l); vals.append(perturb_results[k]['auc'])
             clrs.append(CHART_C['op_colors'][i]); hatches.append(HATCH_OP[i])
             
-    fig, ax = plt.subplots(figsize=(14, 6)); apply_academic_style(fig, ax)
+    fig, ax = plt.subplots(figsize=(16, 7.5)); apply_academic_style(fig, ax)
     bars = ax.bar(range(len(names)), vals, color=clrs, width=0.65, edgecolor='black', linewidth=1.5, zorder=3)
     for bar, h in zip(bars, hatches):
         if h: bar.set_hatch(h)
         
-    for b,v in zip(bars, vals): ax.text(b.get_x()+b.get_width()/2, v+0.01, f'{v:.4f}', ha='center', fontsize=10, fontweight='semibold', color='black')
+    for b,v in zip(bars, vals): ax.text(b.get_x()+b.get_width()/2, v+0.01, f'{v:.4f}', ha='center', fontsize=10.5, fontweight='semibold', color='black')
     ax.axhline(0.5, color='black', ls='--', lw=1.5, label='Random Guess (0.5)', zorder=2)
     ax.axhline(bl_auc, color='black', ls=':', lw=2, label=f'Baseline ({bl_auc:.4f})', zorder=2)
-    ax.set_ylabel('MIA Attack AUC', fontsize=12, fontweight='medium'); ax.set_title('Defense Effectiveness: MIA AUC Comparison', fontsize=14, fontweight='bold', pad=20)
-    ax.set_ylim(0.45, max(vals)+0.05); ax.set_xticks(range(len(names))); ax.set_xticklabels(names, rotation=30, ha='right', fontsize=11)
-    ax.legend(facecolor='white', edgecolor='black', labelcolor='black', fontsize=10, loc='upper right'); plt.tight_layout()
+    ax.set_ylabel('MIA Attack AUC'); ax.set_title('Defense Effectiveness: MIA AUC Comparison', pad=20)
+    ax.set_ylim(0.45, max(vals)+0.05); ax.set_xticks(range(len(names))); ax.set_xticklabels(names, rotation=30, ha='right', fontsize=10.5)
+    ax.legend(facecolor='white', edgecolor='black', labelcolor='black', fontsize=10.5, loc='upper right'); plt.tight_layout()
     return fig
 
 def fig_radar():
@@ -242,7 +249,7 @@ def fig_radar():
     N = len(ms)
     ag = np.linspace(0, 2 * np.pi, N, endpoint=False).tolist() + [0]
 
-    fig, axes = plt.subplots(1, 2, figsize=(CHART_W + 2, 7), subplot_kw=dict(polar=True))
+    fig, axes = plt.subplots(1, 2, figsize=(18, 8), subplot_kw=dict(polar=True))
     fig.patch.set_facecolor('white')
 
     ls_cfgs = [
@@ -262,9 +269,6 @@ def fig_radar():
         ("OP(σ=0.03)", "perturbation_0.03")
     ]
 
-    # 关键修正：两张雷达图共用同一组归一化分母。
-    # 原代码分别在左图和右图内部计算最大值，导致同一个 Baseline 在两张图中的形状不一致。
-    # 这里改为基于所有 LS 与 OP 配置计算全局最大值，使 Baseline 在左右两图中完全一致。
     all_cfgs_for_norm = ls_cfgs + op_cfgs
     global_max = []
     for m_key in mk:
@@ -294,14 +298,14 @@ def fig_radar():
             ax.fill(ag, v, alpha=0.08 if ky == 'baseline' else 0.0, color='black')
 
         ax.set_xticks(ag[:-1])
-        ax.set_xticklabels(ms, fontsize=10, color='black')
+        ax.set_xticklabels(ms, fontsize=10.5, color='black')
         ax.set_yticklabels([])
         ax.set_ylim(0, 1.05)
         ax.set_title(title, fontsize=12, fontweight='700', color='black', pad=18)
         ax.legend(
             loc='upper right',
             bbox_to_anchor=(1.35 if ax_idx == 1 else 1.30, 1.12),
-            fontsize=9,
+            fontsize=10.5,
             framealpha=0.9,
             edgecolor='black'
         )
@@ -317,7 +321,7 @@ def fig_d3_dist_compare():
         ("Label Smoothing (ε=0.2)", "smooth_eps_0.2", None),
         ("Output Perturbation (σ=0.03)", "baseline", 0.03),
     ]
-    fig, axes = plt.subplots(1, 3, figsize=(18, 5.5))
+    fig, axes = plt.subplots(1, 3, figsize=(22, 6.5))
     apply_academic_style(fig, axes)
 
     for idx, (title, key, sigma) in enumerate(configs):
@@ -332,7 +336,6 @@ def fig_d3_dist_compare():
             all_v = np.concatenate([m_losses, nm_losses])
             bins = np.linspace(all_v.min(), all_v.max(), 35)
             
-            # 使用高对比度的底纹
             ax.hist(m_losses, bins=bins, alpha=0.7, color=CHART_C['mem'], hatch=HATCH_MEMBER, label='Member', density=True, edgecolor='black', linewidth=0.8)
             ax.hist(nm_losses, bins=bins, alpha=0.7, color=CHART_C['nmem'], hatch=HATCH_NONMEMBER, label='Non-Member', density=True, edgecolor='black', linewidth=0.8)
             
@@ -341,19 +344,18 @@ def fig_d3_dist_compare():
             ax.axvline(m_mean, color='black', ls='--', lw=2)
             ax.axvline(nm_mean, color='black', ls='-', lw=2)
             ax.annotate(f'Gap={gap:.4f}', xy=((m_mean+nm_mean)/2, ax.get_ylim()[1]*0.85 if ax.get_ylim()[1]>0 else 5),
-                        fontsize=11, fontweight='bold', color='black', ha='center',
+                        fontsize=10.5, fontweight='bold', color='black', ha='center',
                         bbox=dict(boxstyle='round,pad=0.4', fc='white', ec='black', alpha=1.0))
         
-        ax.set_title(title, fontsize=13, fontweight='bold', color='black', pad=15)
-        ax.set_xlabel('Loss', fontsize=12)
-        if idx == 0: ax.set_ylabel('Density', fontsize=12)
-        ax.legend(fontsize=10, facecolor='white', edgecolor='black')
+        ax.set_title(title, pad=15)
+        ax.set_xlabel('Loss')
+        if idx == 0: ax.set_ylabel('Density')
+        ax.legend(fontsize=10.5, facecolor='white', edgecolor='black')
         
-    fig.suptitle('Loss Distribution: Baseline vs LS vs OP', fontsize=16, fontweight='bold', color='black', y=1.05)
+    fig.suptitle('Loss Distribution: Baseline vs LS vs OP', fontsize=14, fontweight='bold', color='black', y=1.02)
     plt.tight_layout(); return fig
 
 def fig_loss_dist():
-    # 仅展示4组标签平滑模型，按“每行2张”排版，避免一行过密
     items = [
         (k, l, gm(k, 'auc'))
         for k, l in zip(LS_KEYS[1:], LS_LABELS_PLOT[1:])
@@ -365,7 +367,7 @@ def fig_loss_dist():
 
     ncols = 2
     nrows = int(np.ceil(n / ncols))
-    fig, axes = plt.subplots(nrows, ncols, figsize=(10, 4.6 * nrows))
+    fig, axes = plt.subplots(nrows, ncols, figsize=(14, 5.5 * nrows))
     axes_flat = np.array(axes).reshape(-1)
     apply_academic_style(fig, axes_flat)
 
@@ -374,22 +376,13 @@ def fig_loss_dist():
         nm = np.array(full_losses[k]['non_member_losses'])
         bins = np.linspace(min(m.min(), nm.min()), max(m.max(), nm.max()), 30)
 
-        # Member 与 Non-Member 使用显著不同的黑白底纹：
-        # Member 为斜线，Non-Member 为点状，图例与柱状图保持一致。
-        ax.hist(
-            m, bins=bins, alpha=0.78, color=CHART_C['mem'], hatch=HATCH_MEMBER,
-            label='Member', density=True, edgecolor='black', linewidth=0.8
-        )
-        ax.hist(
-            nm, bins=bins, alpha=0.78, color=CHART_C['nmem'], hatch=HATCH_NONMEMBER,
-            label='Non-Member', density=True, edgecolor='black', linewidth=0.8
-        )
-        ax.set_title(f'{l}\nAUC={a:.4f}', fontsize=11, fontweight='semibold')
-        ax.set_xlabel('Loss', fontsize=10)
-        ax.set_ylabel('Density', fontsize=10)
-        ax.legend(fontsize=9, facecolor='white', edgecolor='black', labelcolor='black')
+        ax.hist(m, bins=bins, alpha=0.78, color=CHART_C['mem'], hatch=HATCH_MEMBER, label='Member', density=True, edgecolor='black', linewidth=0.8)
+        ax.hist(nm, bins=bins, alpha=0.78, color=CHART_C['nmem'], hatch=HATCH_NONMEMBER, label='Non-Member', density=True, edgecolor='black', linewidth=0.8)
+        ax.set_title(f'{l}\nAUC={a:.4f}')
+        ax.set_xlabel('Loss')
+        ax.set_ylabel('Density')
+        ax.legend(fontsize=10.5, facecolor='white', edgecolor='black', labelcolor='black')
 
-    # 如果子图数量不足，隐藏空白坐标轴
     for ax in axes_flat[n:]:
         ax.axis('off')
 
@@ -402,9 +395,8 @@ def fig_perturb_dist():
     ml = np.array(full_losses['baseline']['member_losses'])
     nl = np.array(full_losses['baseline']['non_member_losses'])
 
-    # 6组输出扰动结果改为3行2列，保证每行两个子图，便于论文排版阅读
     nrows, ncols = 3, 2
-    fig, axes = plt.subplots(nrows, ncols, figsize=(10, 13.5))
+    fig, axes = plt.subplots(nrows, ncols, figsize=(14, 16))
     axes_flat = axes.flatten()
     apply_academic_style(fig, axes_flat)
 
@@ -416,27 +408,19 @@ def fig_perturb_dist():
         v = np.concatenate([mp, np_])
         bins = np.linspace(v.min(), v.max(), 28)
 
-        # 与标签平滑分布图保持一致：Member/ Mem+noise 用斜线，Non/ Non+noise 用点状
-        ax.hist(
-            mp, bins=bins, alpha=0.78, color=CHART_C['mem'], hatch=HATCH_MEMBER,
-            label='Mem+noise', density=True, edgecolor='black', linewidth=0.8
-        )
-        ax.hist(
-            np_, bins=bins, alpha=0.78, color=CHART_C['nmem'], hatch=HATCH_NONMEMBER,
-            label='Non+noise', density=True, edgecolor='black', linewidth=0.8
-        )
-
+        ax.hist(mp, bins=bins, alpha=0.78, color=CHART_C['mem'], hatch=HATCH_MEMBER, label='Mem+noise', density=True, edgecolor='black', linewidth=0.8)
+        ax.hist(np_, bins=bins, alpha=0.78, color=CHART_C['nmem'], hatch=HATCH_NONMEMBER, label='Non+noise', density=True, edgecolor='black', linewidth=0.8)
         pa = gm(f'perturbation_{s}', 'auc')
-        ax.set_title(f'OP(σ={s})\nAUC={pa:.4f}', fontsize=11, fontweight='semibold')
-        ax.set_xlabel('Loss', fontsize=10)
-        ax.set_ylabel('Density', fontsize=10)
-        ax.legend(fontsize=9, facecolor='white', edgecolor='black', labelcolor='black')
+        ax.set_title(f'OP(σ={s})\nAUC={pa:.4f}')
+        ax.set_xlabel('Loss')
+        ax.set_ylabel('Density')
+        ax.legend(fontsize=10.5, facecolor='white', edgecolor='black', labelcolor='black')
 
     plt.tight_layout()
     return fig
 
 def fig_roc_curves():
-    fig, axes = plt.subplots(1, 2, figsize=(16, 7)); apply_academic_style(fig, axes)
+    fig, axes = plt.subplots(1, 2, figsize=(18, 8)); apply_academic_style(fig, axes)
     
     # LS ROC
     ax = axes[0]
@@ -449,7 +433,7 @@ def fig_roc_curves():
         lw = 3.0 if k == 'baseline' else 2.0
         ax.plot(fpr, tpr, color='black', ls=ls_linestyle_cfgs[i], lw=lw, label=f'{l} (AUC={auc_val:.4f})')
     ax.plot([0,1], [0,1], '-', color='gray', lw=1.5, label='Random')
-    ax.set_xlabel('False Positive Rate', fontsize=12, fontweight='medium'); ax.set_ylabel('True Positive Rate', fontsize=12, fontweight='medium'); ax.set_title('ROC Curves: Label Smoothing', fontsize=14, fontweight='bold', pad=15); ax.legend(fontsize=10, facecolor='white', edgecolor='black', labelcolor='black')
+    ax.set_xlabel('False Positive Rate'); ax.set_ylabel('True Positive Rate'); ax.set_title('ROC Curves: Label Smoothing', pad=15); ax.legend(fontsize=10.5, facecolor='white', edgecolor='black', labelcolor='black')
     
     # OP ROC
     ax = axes[1]
@@ -460,11 +444,11 @@ def fig_roc_curves():
             rng_m = np.random.RandomState(42); rng_nm = np.random.RandomState(137); mp = ml_base + rng_m.normal(0, s, len(ml_base)); np_ = nl_base + rng_nm.normal(0, s, len(nl_base)); y_scores_p = np.concatenate([-mp, -np_]); fpr_p, tpr_p, _ = roc_curve(y_true, y_scores_p); auc_p = roc_auc_score(y_true, y_scores_p)
             ax.plot(fpr_p, tpr_p, color='black', ls=OP_LINESTYLES[i % len(OP_LINESTYLES)], lw=1.5, label=f'OP(σ={s}) (AUC={auc_p:.4f})')
     ax.plot([0,1], [0,1], '-', color='gray', lw=1.5, label='Random')
-    ax.set_xlabel('False Positive Rate', fontsize=12, fontweight='medium'); ax.set_ylabel('True Positive Rate', fontsize=12, fontweight='medium'); ax.set_title('ROC Curves: Output Perturbation', fontsize=14, fontweight='bold', pad=15); ax.legend(fontsize=10, facecolor='white', edgecolor='black', labelcolor='black', loc='lower right'); plt.tight_layout()
+    ax.set_xlabel('False Positive Rate'); ax.set_ylabel('True Positive Rate'); ax.set_title('ROC Curves: Output Perturbation', pad=15); ax.legend(fontsize=10.5, facecolor='white', edgecolor='black', labelcolor='black', loc='lower right'); plt.tight_layout()
     return fig
 
 def fig_tpr_at_low_fpr():
-    fig, axes = plt.subplots(1, 2, figsize=(16, 6.5)); apply_academic_style(fig, axes); labels_all, tpr5_all, tpr1_all, clrs_all, hatches_all = [], [], [], [], []
+    fig, axes = plt.subplots(1, 2, figsize=(18, 7.5)); apply_academic_style(fig, axes); labels_all, tpr5_all, tpr1_all, clrs_all, hatches_all = [], [], [], [], []
     ls_h_list = [HATCH_BASELINE] + HATCH_LS
     ls_c_list = [CHART_C['baseline']] + CHART_C['ls_colors']
     
@@ -479,19 +463,19 @@ def fig_tpr_at_low_fpr():
     bars = ax.bar(x, tpr5_all, color=clrs_all, width=0.65, edgecolor='black', linewidth=1.5, zorder=3)
     for bar, h in zip(bars, hatches_all):
         if h: bar.set_hatch(h)
-    for b, v in zip(bars, tpr5_all): ax.text(b.get_x()+b.get_width()/2, v+0.005, f'{v:.3f}', ha='center', fontsize=9, fontweight='semibold', color='black')
-    ax.set_ylabel('TPR @ 5% FPR', fontsize=12, fontweight='medium'); ax.set_title('Attack Power at 5% FPR', fontsize=14, fontweight='bold', pad=15); ax.set_xticks(x); ax.set_xticklabels(labels_all, rotation=35, ha='right', fontsize=11); ax.axhline(0.05, color='gray', ls='--', lw=2, label='Random (0.05)'); ax.legend(facecolor='white', edgecolor='black', labelcolor='black', fontsize=10)
+    for b, v in zip(bars, tpr5_all): ax.text(b.get_x()+b.get_width()/2, v+0.005, f'{v:.3f}', ha='center', fontsize=10.5, fontweight='semibold', color='black')
+    ax.set_ylabel('TPR @ 5% FPR'); ax.set_title('Attack Power at 5% FPR', pad=15); ax.set_xticks(x); ax.set_xticklabels(labels_all, rotation=35, ha='right', fontsize=10.5); ax.axhline(0.05, color='gray', ls='--', lw=2, label='Random (0.05)'); ax.legend(facecolor='white', edgecolor='black', labelcolor='black', fontsize=10.5)
     
     ax = axes[1]; 
     bars = ax.bar(x, tpr1_all, color=clrs_all, width=0.65, edgecolor='black', linewidth=1.5, zorder=3)
     for bar, h in zip(bars, hatches_all):
         if h: bar.set_hatch(h)
-    for b, v in zip(bars, tpr1_all): ax.text(b.get_x()+b.get_width()/2, v+0.003, f'{v:.3f}', ha='center', fontsize=9, fontweight='semibold', color='black')
-    ax.set_ylabel('TPR @ 1% FPR', fontsize=12, fontweight='medium'); ax.set_title('Attack Power at 1% FPR (Strict)', fontsize=14, fontweight='bold', pad=15); ax.set_xticks(x); ax.set_xticklabels(labels_all, rotation=35, ha='right', fontsize=11); ax.axhline(0.01, color='gray', ls='--', lw=2, label='Random (0.01)'); ax.legend(facecolor='white', edgecolor='black', labelcolor='black', fontsize=10); plt.tight_layout()
+    for b, v in zip(bars, tpr1_all): ax.text(b.get_x()+b.get_width()/2, v+0.003, f'{v:.3f}', ha='center', fontsize=10.5, fontweight='semibold', color='black')
+    ax.set_ylabel('TPR @ 1% FPR'); ax.set_title('Attack Power at 1% FPR (Strict)', pad=15); ax.set_xticks(x); ax.set_xticklabels(labels_all, rotation=35, ha='right', fontsize=10.5); ax.axhline(0.01, color='gray', ls='--', lw=2, label='Random (0.01)'); ax.legend(facecolor='white', edgecolor='black', labelcolor='black', fontsize=10.5); plt.tight_layout()
     return fig
 
 def fig_loss_gap_waterfall():
-    fig, ax = plt.subplots(figsize=(14, 6)); apply_academic_style(fig, ax); names, gaps, clrs, hatches = [], [], [], []
+    fig, ax = plt.subplots(figsize=(16, 7.5)); apply_academic_style(fig, ax); names, gaps, clrs, hatches = [], [], [], []
     ls_h_list = [HATCH_BASELINE] + HATCH_LS
     ls_c_list = [CHART_C['baseline']] + CHART_C['ls_colors']
     for i, (k, l) in enumerate(zip(LS_KEYS, LS_LABELS_PLOT)): 
@@ -504,8 +488,8 @@ def fig_loss_gap_waterfall():
     bars = ax.bar(range(len(names)), gaps, color=clrs, width=0.65, edgecolor='black', linewidth=1.5, zorder=3)
     for bar, h in zip(bars, hatches):
         if h: bar.set_hatch(h)
-    for b, v in zip(bars, gaps): ax.text(b.get_x()+b.get_width()/2, v+0.0005, f'{v:.4f}', ha='center', fontsize=10, fontweight='semibold', color='black')
-    ax.set_ylabel('Loss Gap', fontsize=12, fontweight='medium'); ax.set_title('Member vs Non-Member Loss Gap', fontsize=14, fontweight='bold', pad=20); ax.set_xticks(range(len(names))); ax.set_xticklabels(names, rotation=30, ha='right', fontsize=11); ax.annotate('Smaller gap = Better Privacy', xy=(8, gaps[0]*0.4), fontsize=11, color='black', fontstyle='italic', ha='center', backgroundcolor='white', bbox=dict(boxstyle='round,pad=0.4', facecolor='white', edgecolor='black', alpha=1.0)); plt.tight_layout()
+    for b, v in zip(bars, gaps): ax.text(b.get_x()+b.get_width()/2, v+0.0005, f'{v:.4f}', ha='center', fontsize=10.5, fontweight='semibold', color='black')
+    ax.set_ylabel('Loss Gap'); ax.set_title('Member vs Non-Member Loss Gap', pad=20); ax.set_xticks(range(len(names))); ax.set_xticklabels(names, rotation=30, ha='right', fontsize=10.5); ax.annotate('Smaller gap = Better Privacy', xy=(8, gaps[0]*0.4), fontsize=10.5, color='black', fontstyle='italic', ha='center', backgroundcolor='white', bbox=dict(boxstyle='round,pad=0.4', facecolor='white', edgecolor='black', alpha=1.0)); plt.tight_layout()
     return fig
 
 def fig_acc_bar():
@@ -521,52 +505,52 @@ def fig_acc_bar():
             names.append(l); vals.append(bl_acc)
             clrs.append(CHART_C['op_colors'][i]); hatches.append(HATCH_OP[i])
         
-    fig, ax = plt.subplots(figsize=(12, 7)); apply_academic_style(fig, ax)
+    fig, ax = plt.subplots(figsize=(14, 8)); apply_academic_style(fig, ax)
     bars = ax.bar(range(len(names)), vals, color=clrs, width=0.65, edgecolor='black', linewidth=1.5, zorder=3)
     for bar, h in zip(bars, hatches):
         if h: bar.set_hatch(h)
-    for b, v in zip(bars, vals): ax.text(b.get_x()+b.get_width()/2, v+1, f'{v:.1f}%', ha='center', fontsize=11, fontweight='bold', color='black')
-    ax.set_ylabel('Test Accuracy (%)', fontsize=12, fontweight='medium'); ax.set_title('Model Utility: Test Accuracy', fontsize=15, fontweight='bold', pad=20)
-    ax.set_ylim(0, 105); ax.set_xticks(range(len(names))); ax.set_xticklabels(names, rotation=35, ha='right', fontsize=12); plt.tight_layout()
+    for b, v in zip(bars, vals): ax.text(b.get_x()+b.get_width()/2, v+1, f'{v:.1f}%', ha='center', fontsize=10.5, fontweight='bold', color='black')
+    ax.set_ylabel('Test Accuracy (%)'); ax.set_title('Model Utility: Test Accuracy', pad=20)
+    ax.set_ylim(0, 105); ax.set_xticks(range(len(names))); ax.set_xticklabels(names, rotation=35, ha='right', fontsize=10.5); plt.tight_layout()
     return fig
 
 def fig_tradeoff():
-    fig, ax = plt.subplots(figsize=(12, 7)); apply_academic_style(fig, ax); 
+    fig, ax = plt.subplots(figsize=(14, 8)); apply_academic_style(fig, ax); 
     markers_ls = ['o', 's', 'p', '*', 'h']
     for i, (k, l) in enumerate(zip(LS_KEYS, LS_LABELS_PLOT)):
         if k in mia_results and k in utility_results: 
-            ax.scatter(utility_results[k]['accuracy']*100, mia_results[k]['auc'], label=l, marker=markers_ls[i], color='white', s=250, edgecolors='black', lw=2.0, zorder=5)
+            ax.scatter(utility_results[k]['accuracy']*100, mia_results[k]['auc'], label=l, marker=markers_ls[i], color='white', s=280, edgecolors='black', lw=2.0, zorder=5)
     op_markers = ['^', 'D', 'v', 'P', 'X', '>']
     for i, (k, l) in enumerate(zip(OP_KEYS, OP_LABELS_PLOT)):
         if k in perturb_results: 
-            ax.scatter(bl_acc, perturb_results[k]['auc'], label=l, marker=op_markers[i], color='#AAAAAA', s=200, edgecolors='black', lw=1.5, zorder=6)
+            ax.scatter(bl_acc, perturb_results[k]['auc'], label=l, marker=op_markers[i], color='#AAAAAA', s=230, edgecolors='black', lw=1.5, zorder=6)
     
     ax.axhline(0.5, color='black', ls='--', lw=1.5, alpha=0.6, label='Random (AUC=0.5)')
-    ax.annotate('IDEAL ZONE\nHigh Utility, Low Risk', xy=(85, 0.51), fontsize=11, fontweight='bold', color='black', ha='center', bbox=dict(boxstyle='round,pad=0.5', fc='white', ec='black'))
-    ax.annotate('HIGH RISK ZONE\nLow Utility, High Risk', xy=(62, 0.61), fontsize=11, fontweight='bold', color='black', ha='center', bbox=dict(boxstyle='round,pad=0.5', fc='white', ec='black'))
-    ax.set_xlabel('Model Utility (Accuracy %)', fontsize=12, fontweight='medium'); ax.set_ylabel('Privacy Risk (MIA AUC)', fontsize=12, fontweight='medium')
-    ax.set_title('Privacy-Utility Trade-off Analysis', fontsize=15, fontweight='bold', pad=20)
-    ax.legend(fontsize=11, loc='lower left', ncol=2, facecolor='white', edgecolor='black', labelcolor='black'); plt.tight_layout()
+    ax.annotate('IDEAL ZONE\nHigh Utility, Low Risk', xy=(85, 0.51), fontsize=10.5, fontweight='bold', color='black', ha='center', bbox=dict(boxstyle='round,pad=0.5', fc='white', ec='black'))
+    ax.annotate('HIGH RISK ZONE\nLow Utility, High Risk', xy=(62, 0.61), fontsize=10.5, fontweight='bold', color='black', ha='center', bbox=dict(boxstyle='round,pad=0.5', fc='white', ec='black'))
+    ax.set_xlabel('Model Utility (Accuracy %)'); ax.set_ylabel('Privacy Risk (MIA AUC)')
+    ax.set_title('Privacy-Utility Trade-off Analysis', pad=20)
+    ax.legend(fontsize=10.5, loc='lower left', ncol=2, facecolor='white', edgecolor='black', labelcolor='black'); plt.tight_layout()
     return fig
 
 def fig_auc_trend():
-    fig, axes = plt.subplots(1, 2, figsize=(16, 6.5)); apply_academic_style(fig, axes); ax = axes[0]; eps_vals = [0.0, 0.02, 0.05, 0.1, 0.2]; auc_vals = [gm(k, 'auc') for k in LS_KEYS]; acc_vals = [gu(k) for k in LS_KEYS]
+    fig, axes = plt.subplots(1, 2, figsize=(18, 7.5)); apply_academic_style(fig, axes); ax = axes[0]; eps_vals = [0.0, 0.02, 0.05, 0.1, 0.2]; auc_vals = [gm(k, 'auc') for k in LS_KEYS]; acc_vals = [gu(k) for k in LS_KEYS]
     ax2 = ax.twinx(); 
     line1 = ax.plot(eps_vals, auc_vals, marker='o', ls='-', color='black', lw=3, ms=9, label='MIA AUC (Risk)', zorder=5); 
     line2 = ax2.plot(eps_vals, acc_vals, marker='s', ls='--', color='black', lw=3, ms=9, label='Utility % (right)', zorder=5); 
     ax.axhline(0.5, color='gray', ls=':')
     ax.fill_between(eps_vals, auc_vals, 0.5, alpha=0.2, color='gray', hatch='//')
-    ax.set_xlabel('Label Smoothing ε', fontsize=12, fontweight='medium'); ax.set_ylabel('MIA AUC', fontsize=12, fontweight='medium', color='black'); ax2.set_ylabel('Utility (%)', fontsize=12, fontweight='medium', color='black'); ax.set_title('Label Smoothing Trends', fontsize=14, fontweight='bold', pad=15); ax.tick_params(axis='y', labelcolor='black'); ax2.tick_params(axis='y', labelcolor='black'); ax2.spines['right'].set_color('black'); ax2.spines['left'].set_color('black'); lines = line1 + line2; labels = [l.get_label() for l in lines]
-    ax.legend(lines, labels, fontsize=10, facecolor='white', edgecolor='black', loc='lower right')
+    ax.set_xlabel('Label Smoothing ε'); ax.set_ylabel('MIA AUC', color='black'); ax2.set_ylabel('Utility (%)', color='black'); ax.set_title('Label Smoothing Trends', pad=15); ax.tick_params(axis='y', labelcolor='black'); ax2.tick_params(axis='y', labelcolor='black'); ax2.spines['right'].set_color('black'); ax2.spines['left'].set_color('black'); lines = line1 + line2; labels = [l.get_label() for l in lines]
+    ax.legend(lines, labels, fontsize=10.5, facecolor='white', edgecolor='black', loc='lower right')
     
     ax = axes[1]; sig_vals = OP_SIGMAS; auc_op = [gm(k, 'auc') for k in OP_KEYS]; 
     ax.plot(sig_vals, auc_op, marker='^', ls='-', color='black', lw=3, ms=9, zorder=5, label='MIA AUC'); 
     ax.axhline(bl_auc, color='black', ls='--', lw=2, label=f'Baseline ({bl_auc:.4f})'); 
     ax.axhline(0.5, color='gray', ls=':', label='Random (0.5)'); 
     ax.fill_between(sig_vals, auc_op, bl_auc, alpha=0.2, color='gray', hatch='\\\\', label='AUC Reduction')
-    ax2r = ax.twinx(); ax2r.axhline(bl_acc, color='black', ls='-', lw=2.5); ax2r.set_ylabel(f'Utility = {bl_acc:.1f}% (unchanged)', fontsize=12, fontweight='medium', color='black'); ax2r.set_ylim(0,100); ax2r.tick_params(axis='y', labelcolor='black'); ax2r.spines['right'].set_color('black')
-    ax.set_xlabel('Perturbation σ', fontsize=12, fontweight='medium'); ax.set_ylabel('MIA AUC', fontsize=12, fontweight='medium'); ax.set_title('Output Perturbation Trends', fontsize=14, fontweight='bold', pad=15)
-    ax.legend(fontsize=10, facecolor='white', edgecolor='black', loc='lower left'); plt.tight_layout()
+    ax2r = ax.twinx(); ax2r.axhline(bl_acc, color='black', ls='-', lw=2.5); ax2r.set_ylabel(f'Utility = {bl_acc:.1f}% (unchanged)', color='black'); ax2r.set_ylim(0,100); ax2r.tick_params(axis='y', labelcolor='black'); ax2r.spines['right'].set_color('black')
+    ax.set_xlabel('Perturbation σ'); ax.set_ylabel('MIA AUC'); ax.set_title('Output Perturbation Trends', pad=15)
+    ax.legend(fontsize=10.5, facecolor='white', edgecolor='black', loc='lower left'); plt.tight_layout()
     return fig
 
 # ================================================================
@@ -765,7 +749,6 @@ footer { display: none !important; }
 # ================================================================
 # UI 布局构建 (完全不碰原版Blocks构建)
 # ================================================================
-# 移除了警告的 theme 和 css 参数，确保兼容 Gradio 6.0
 with gr.Blocks(title="MIA攻防研究") as demo:
     
     gr.HTML("""<div class="title-area">
@@ -1045,5 +1028,4 @@ with gr.Blocks(title="MIA攻防研究") as demo:
 
 """)
 
-# 添加了 theme 和 css 并修复了括号问题
 demo.launch(theme=gr.themes.Soft(), css=CSS)