add models

app.py

@@ -9,7 +9,6 @@ import gradio as gr
 
 RESULTS_DIR = "./worldlens-results"
 
-# 指标好坏方向
 METRICS_MIN_BETTER = [
     "Depth Discrepancy", "Perceptual Discrepancy",
     "Photometric Error", "Geometric Discrepancy",
@@ -20,7 +19,7 @@ METRICS_MIN_BETTER = [
 METRICS_MAX_BETTER = [
     "Subject Fidelity", "Subject Coherence", "Subject Consistency",
     "Temporal Consistency", "Semantic Consistency",
-    "View Consistency",             # 你的 JSON 里有这个，默认认为越大越好
+    "View Consistency",
     "Novel-View Quality",
     "Open-Loop Adherence", "Route Completion", "Closed-Loop Adherence",
     "Map Segmentation", "3D Object Detection", "3D Object Tracking",
@@ -32,19 +31,14 @@ METRIC_BETTER: Dict[str, Literal["min", "max"]] = {
 }
 METRIC_BETTER.update({m: "max" for m in METRICS_MAX_BETTER})
 
-# 下拉框展示的所有指标（去重+排序）
 METRIC_CHOICES: List[str] = sorted(set(METRICS_MIN_BETTER + METRICS_MAX_BETTER))
 DEFAULT_METRIC = "Subject Fidelity" if "Subject Fidelity" in METRIC_CHOICES else METRIC_CHOICES[0]
 
-# 全局 DataFrame（所有模型）
 df_all: Optional[pd.DataFrame] = None
 
 
 def load_results() -> pd.DataFrame:
-    """
-    从 ./worldlens-results 读取所有 json，整理成一个宽表：
-    每一行是一个模型，每一列是一个指标。
-    """
+
     rows = []
 
     json_files = sorted(glob.glob(os.path.join(RESULTS_DIR, "*.json")))
@@ -57,7 +51,7 @@ def load_results() -> pd.DataFrame:
 
         model_name = os.path.splitext(os.path.basename(path))[0]
         venue = data.get("venue", "")
-        date = data.get("data", "")  # 你这边字段叫 data，我就直接用
+        date = data.get("date", "")
 
         row = {
             "Model": model_name,
@@ -66,7 +60,6 @@ def load_results() -> pd.DataFrame:
         }
 
         metrics = data.get("Metrics", {})
-        # 展开所有子字典，列名直接用 metric 名称（假设唯一）
         for category, metric_dict in metrics.items():
             if not isinstance(metric_dict, dict):
                 continue
@@ -77,7 +70,6 @@ def load_results() -> pd.DataFrame:
 
     df = pd.DataFrame(rows)
 
-    # 统一列顺序：meta + 指标
     meta_cols = ["Model", "venue", "date"]
     metric_cols = [c for c in df.columns if c not in meta_cols]
     df = df[meta_cols + metric_cols]
@@ -100,15 +92,10 @@ def update_leaderboard(
     sort_mode: str,
     selected_metrics: Optional[List[str]],
 ) -> Tuple[pd.DataFrame, plt.Figure]:
-    """
-    根据用户选择更新排行榜表格与条形图。
-    metric: 用于排序 & 画图的主指标
-    selected_metrics: 勾选的“想在表格中展示”的其它指标（可以多个）
-    """
+
     global df_all
 
     if df_all is None or df_all.empty:
-        # 空表兜底
         fig, ax = plt.subplots(figsize=(6, 3))
         ax.text(0.5, 0.5, "No results found in ./worldlens-results",
                 ha="center", va="center")
@@ -117,11 +104,9 @@ def update_leaderboard(
 
     df = df_all.copy()
 
-    # 模型名过滤
     if model_filter:
         df = df[df["Model"].str.contains(model_filter, case=False, regex=False)]
 
-    # venue 过滤
     if venue_filter and venue_filter != "All":
         df = df[df["venue"] == venue_filter]
 
@@ -131,30 +116,22 @@ def update_leaderboard(
         ax.axis("off")
         return pd.DataFrame(), fig
 
-    # 排序方向
     better = METRIC_BETTER.get(metric, "max")
     if sort_mode == "Auto":
         ascending = (better == "min")
     elif sort_mode == "Ascending (small → large)":
         ascending = True
-    else:  # "Descending (large → small)"
+    else:
         ascending = False
 
     df_sorted = df.sort_values(metric, ascending=ascending)
 
-    # Top-K
     df_top = df_sorted.head(top_k).copy()
-
-    # 构造表格列：
-    # 固定: Model, venue, date
-    # + 勾选的指标
-    # + 排序指标（如果没选）
     cols = ["Model", "venue", "date"]
 
     if selected_metrics is None:
         selected_metrics = []
 
-    # 去掉不在 df_top 里的指标（有些 metric 可能某些 json 里没计算）
     for m in selected_metrics:
         if m in df_top.columns and m not in cols:
             cols.append(m)
@@ -164,30 +141,63 @@ def update_leaderboard(
 
     table_df = df_top[cols].round(3)
 
-    # 画条形图（只画排序指标）
-    fig, ax = plt.subplots(figsize=(9, 4))
-    ax.barh(table_df["Model"], df_top[metric].iloc[:len(table_df)])
-    ax.set_xlabel(metric)
-    ax.set_ylabel("Model")
-    ax.set_title(f"Leaderboard by {metric}")
+    # =========================
+    # Dark-theme leaderboard plot
+    # =========================
+    bg_color = "#0e1117"        # HF 深色背景
+    panel_color = "#161b22"     # 面板背景
+    bar_color = "#4cc9f0"       # 主色（青蓝）
+    grid_color = "#30363d"
+    text_color = "#c9d1d9"
+
+    fig, ax = plt.subplots(figsize=(10, 4.5))
+    fig.patch.set_facecolor(bg_color)
+    ax.set_facecolor(panel_color)
+
+    values = df_top[metric].values
+    models = table_df["Model"].values
+    bars = ax.barh(models, values, color=bar_color, height=0.6)
 
-    # 为了让「最好的」在上面：如果按升序(小→大)，我们反转 y 轴，让更小的在上。
     if ascending:
         ax.invert_yaxis()
 
+    ax.set_xlabel(metric, color=text_color, fontsize=11, labelpad=6)
+    ax.set_title(
+        f"Leaderboard · {metric}",
+        fontsize=13,
+        color=text_color,
+        pad=10,
+        fontweight="bold"
+    )
+
+    ax.xaxis.grid(True, linestyle="--", linewidth=0.6, color=grid_color, alpha=0.7)
+    ax.yaxis.grid(False)
+
+    for spine in ["top", "right", "left"]:
+        ax.spines[spine].set_visible(False)
+    ax.spines["bottom"].set_color(grid_color)
+
+    ax.tick_params(axis="x", colors=text_color, labelsize=10)
+    ax.tick_params(axis="y", colors=text_color, labelsize=10)
+
+    for bar, value in zip(bars, values):
+        ax.text(
+            bar.get_width() * 1.01,
+            bar.get_y() + bar.get_height() / 2,
+            f"{value:.2f}",
+            va="center",
+            ha="left",
+            fontsize=9.5,
+            color=text_color
+        )
+
     plt.tight_layout()
 
     return table_df, fig
 
 
 def reload_data():
-    """
-    点击“Reload JSONs” / 页面加载时调用：
-    重新加载所有 json，并返回：
-      - 状态文字
-      - venue_dropdown 的更新
-      - 默认的表格和图
-    """
+
     global df_all
     df_all = load_results()
 
@@ -204,7 +214,6 @@ def reload_data():
     venue_choices = get_venue_choices(df_all)
     msg = f"Loaded {len(df_all)} models from {RESULTS_DIR}"
 
-    # 用默认 metric 画一次（selected_metrics 先用一个简单默认）
     default_selected = ["Subject Fidelity", "Temporal Consistency", "Map Segmentation"]
     default_selected = [m for m in default_selected if m in METRIC_CHOICES]
 
@@ -234,7 +243,6 @@ with gr.Blocks(css="""
     gr.Markdown(
         """
 # 🌍 WorldLens Leaderboard
-> from WorldBench Team
         """,
         elem_id="title"
     )
@@ -243,13 +251,13 @@ with gr.Blocks(css="""
 
     with gr.Row():
         metric_dropdown = gr.Dropdown(
-            label="排序指标 / Metric (for ranking)",
+            label="Metric (for ranking)",
             choices=METRIC_CHOICES,           # 固定 choices，避免动态更新不兼容
             value=DEFAULT_METRIC,
             interactive=True,
         )
         sort_mode_radio = gr.Radio(
-            label="排序方式 / Sort mode",
+            label="Sort mode",
             choices=[
                 "Auto",
                 "Ascending (small → large)",
@@ -259,7 +267,7 @@ with gr.Blocks(css="""
             interactive=True,
         )
         topk_slider = gr.Slider(
-            label="显示 Top-K 模型 / Top-K",
+            label="Top-K",
             minimum=3,
             maximum=50,
             value=10,
@@ -269,7 +277,7 @@ with gr.Blocks(css="""
 
     # 新增：表格中展示的多个指标
     metrics_select = gr.CheckboxGroup(
-        label="在表格中一起展示的指标 / Metrics to show in table",
+        label="Metrics to show in table",
         choices=METRIC_CHOICES,
         value=["Subject Fidelity", "Temporal Consistency", "Map Segmentation"],
         interactive=True,
@@ -277,12 +285,12 @@ with gr.Blocks(css="""
 
     with gr.Row():
         model_filter_box = gr.Textbox(
-            label="模型名过滤（包含关系）/ Filter by model name",
-            placeholder="例如: magic, dream, ...",
+            label="Filter by model name",
+            placeholder="magic, dream, ...",
             interactive=True,
         )
         venue_dropdown = gr.Dropdown(
-            label="按 Venue 筛选 / Filter by venue",
+            label="Filter by venue",
             choices=["All"],
             value="All",
             interactive=True,
@@ -296,17 +304,14 @@ with gr.Blocks(css="""
         label="Leaderboard",
         interactive=False,
     )
-    # 显式指定 format="png"，避免 webp 不支持的问题
     leaderboard_plot = gr.Plot(label="Metric comparison", format="png")
 
-    # 点击 Reload：重新加载 + 更新 venue + 表格与图
     reload_button.click(
         fn=reload_data,
         inputs=[],
         outputs=[status_box, venue_dropdown, leaderboard_table, leaderboard_plot],
     )
 
-    # 更新排行榜（多传一个 selected_metrics）
     update_button.click(
         fn=update_leaderboard,
         inputs=[
@@ -320,7 +325,6 @@ with gr.Blocks(css="""
         outputs=[leaderboard_table, leaderboard_plot],
     )
 
-    # 页面加载时自动尝试加载一次
     demo.load(
         fn=reload_data,
         inputs=[],
@@ -329,4 +333,4 @@ with gr.Blocks(css="""
 
 
 if __name__ == "__main__":
-    demo.launch()  # 本地想公网访问可以改成 demo.launch(share=True)
+    demo.launch()

worldlens-results/DrivingSphere.json

@@ -0,0 +1,12 @@
+{
+    "venue": "CVPR'2025",
+    "date": "2024-11-18",
+    "Metrics": {
+        "Action-Following":{
+            "Displacement Error": 0.54,
+            "Open-Loop Adherence": 76.02,
+            "Route Completion": 11.02,
+            "Closed-Loop Adherence": 8.29
+        }
+    }
+}

worldlens-results/Panacea.json

@@ -0,0 +1,9 @@
+{
+    "venue": "CVPR'2024",
+    "date": "2023-11-28",
+    "Metrics": {
+        "Action-Following":{
+            "Displacement Error": 0.58
+        }
+    }
+}

worldlens-results/RLGF.json

@@ -0,0 +1,12 @@
+{
+    "venue": "NeurlPS'2025",
+    "date": "2025-09-20",
+    "Metrics": {
+        "Action-Following":{
+            "Displacement Error": 0.53,
+            "Open-Loop Adherence": 78.45,
+            "Route Completion": 13.51,
+            "Closed-Loop Adherence": 10.59
+        }
+    }
+}

worldlens-results/dist4d.json

@@ -0,0 +1,28 @@
+{
+    "venue": "ICCV'2025",
+    "date": "2025-03-19",
+    "Metrics": {
+        "Generation": {
+            "Subject Fidelity": 30.32,
+            "Subject Coherence": 79.36,
+            "Subject Consistency": 74.69,
+            "Depth Discrepancy": 17.71,
+            "Temporal Consistency": 77.76,
+            "Semantic Consistency": 84.32,
+            "Perceptual Discrepancy": 58.08,
+            "View Consistency": 389.78
+        },
+        "Reconstruction":{
+            "Photometric Error": 0.066,
+            "Geometric Discrepancy": 0.080,
+            "Novel-View Quality": 43.09,
+            "Novel-View Discrepancy": 192.39
+        },
+        "Downstream Task":{
+            "Map Segmentation": 35.55,
+            "3D Object Detection": 33.22,
+            "3D Object Tracking": 15.30,
+            "Occupancy Prediction": 26.10
+        }
+    }
+}

worldlens-results/dreamforge.json

@@ -1,34 +1,34 @@
 {
     "venue": "Arxiv'2024",
-    "date": "2024-10-04",
+    "date": "2024-09-06",
     "Metrics": {
         "Generation": {
-            "Subject Fidelity": 28.49,
-            "Subject Coherence": 75.95,
-            "Subject Consistency": 65.22,
-            "Depth Discrepancy": 24.19,
-            "Temporal Consistency": 74.44,
-            "Semantic Consistency": 80.63,
-            "Perceptual Discrepancy": 222.00,
-            "View Consistency": 185.77
+            "Subject Fidelity": 31.99,
+            "Subject Coherence": 75.12,
+            "Subject Consistency": 76.40,
+            "Depth Discrepancy": 19.27,
+            "Temporal Consistency": 79.82,
+            "Semantic Consistency": 84.99,
+            "Perceptual Discrepancy": 189.76,
+            "View Consistency": 194.99
         },
         "Reconstruction":{
-            "Photometric Error": 0.140,
-            "Geometric Discrepancy": 0.115,
-            "Novel-View Quality": 39.82,
-            "Novel-View Discrepancy": 427.30
+            "Photometric Error": 0.097,
+            "Geometric Discrepancy": 0.105,
+            "Novel-View Quality": 41.23,
+            "Novel-View Discrepancy": 347.70
         },
         "Action-Following":{
-            "Displacement Error": 0.57,
-            "Open-Loop Adherence": 71.23,
-            "Route Completion": 6.89,
-            "Closed-Loop Adherence": 4.82
+            "Displacement Error": 0.55,
+            "Open-Loop Adherence": 75.51,
+            "Route Completion": 10.23,
+            "Closed-Loop Adherence": 7.65
         },
         "Downstream Task":{
-            "Map Segmentation": 18.34,
-            "3D Object Detection": 22.41,
-            "3D Object Tracking": 7.90,
-            "Occupancy Prediction": 23.14
+            "Map Segmentation": 30.91,
+            "3D Object Detection": 26.71,
+            "3D Object Tracking": 10.30,
+            "Occupancy Prediction": 23.71
         }
     }
 }

worldlens-results/drivedreamer2.json

@@ -0,0 +1,28 @@
+{
+    "venue": "AAAI'2025",
+    "date": "2024-03-11",
+    "Metrics": {
+        "Generation": {
+            "Subject Fidelity": 27.38,
+            "Subject Coherence": 78.97,
+            "Subject Consistency": 74.49,
+            "Depth Discrepancy": 17.73,
+            "Temporal Consistency": 79.51,
+            "Semantic Consistency": 85.91,
+            "Perceptual Discrepancy": 127.07,
+            "View Consistency": 302.83
+        },
+        "Reconstruction":{
+            "Photometric Error": 0.093,
+            "Geometric Discrepancy": 0.073,
+            "Novel-View Quality": 36.10,
+            "Novel-View Discrepancy": 259.91
+        },
+        "Downstream Task":{
+            "Map Segmentation": 33.62,
+            "3D Object Detection": 30.90,
+            "3D Object Tracking": 13.30,
+            "Occupancy Prediction": 26.82
+        }
+    }
+}

worldlens-results/magicdriveV2.json

@@ -0,0 +1,12 @@
+{
+    "venue": "ICCV'2025",
+    "date": "2024-11-21",
+    "Metrics": {
+        "Action-Following":{
+            "Displacement Error": 0.53,
+            "Open-Loop Adherence": 78.91,
+            "Route Completion": 12.31,
+            "Closed-Loop Adherence": 9.50
+        }
+    }
+}

worldlens-results/opendwm.json

@@ -0,0 +1,28 @@
+{
+    "venue": "CVPR'2025",
+    "date": "2025-01-09",
+    "Metrics": {
+        "Generation": {
+            "Subject Fidelity": 36.30,
+            "Subject Coherence": 83.13,
+            "Subject Consistency": 78.33,
+            "Depth Discrepancy": 18.17,
+            "Temporal Consistency": 79.63,
+            "Semantic Consistency": 84.08,
+            "Perceptual Discrepancy": 90.42,
+            "View Consistency": 211.18
+        },
+        "Reconstruction":{
+            "Photometric Error": 0.065,
+            "Geometric Discrepancy": 0.080,
+            "Novel-View Quality": 43.09,
+            "Novel-View Discrepancy": 192.39
+        },
+        "Downstream Task":{
+            "Map Segmentation": 27.63,
+            "3D Object Detection": 21.96,
+            "3D Object Tracking": 6.90,
+            "Occupancy Prediction": 24.82
+        }
+    }
+}

worldlens-results/xscene.json

@@ -0,0 +1,28 @@
+{
+    "venue": "NerulPS'2025",
+    "date": "2025-06-16",
+    "Metrics": {
+        "Generation": {
+            "Subject Fidelity": 27.17,
+            "Subject Coherence": 77.22,
+            "Subject Consistency": 74.37,
+            "Depth Discrepancy": 20.50,
+            "Temporal Consistency": 79.41,
+            "Semantic Consistency": 83.80,
+            "Perceptual Discrepancy": 179.74,
+            "View Consistency": 201.00
+        },
+        "Reconstruction":{
+            "Photometric Error": 0.098,
+            "Geometric Discrepancy": 0.096,
+            "Novel-View Quality": 38.04,
+            "Novel-View Discrepancy": 365.71
+        },
+        "Downstream Task":{
+            "Map Segmentation": 27.27,
+            "3D Object Detection": 29.89,
+            "3D Object Tracking": 8.80,
+            "Occupancy Prediction": 23.68
+        }
+    }
+}