Update app.py

app.py

@@ -58,7 +58,7 @@ LEADERBOARD_DF = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, COLS,
 
 # --- 📌 Define Column Groups ---
 
-# Tab 1: Generation (Strictly matching Table 1 - 9 columns + Model)
+# Tab 1: Generation (Table 1)
 COLS_GEN = [
     "Model",
     # LongText2Video
@@ -69,19 +69,18 @@ COLS_GEN = [
     "Video2Video / CLIP", "Video2Video / DINO", "Video2Video / MLLM"
 ]
 
-# Tab 2: Long-Video Tasks (Understanding, Editing, Segmentation)
+# Tab 2: Perception & Editing (Table 2 - 7 Data Columns)
 COLS_LONG = [
     "Model", 
     "LongVideo QA / Acc",      # Understanding
-    "Editing / MLLM",          # Editing
-    "Segmentation / J&F"       # Segmentation
+    "Editing / CLIP",          # Editing
+    "Editing / DINO",
+    "Editing / MLLM",
+    "Segmentation / J",        # Segmentation
+    "Segmentation / F",
+    "Segmentation / J&F"       
 ]
 
-# Filter out columns not present in the dataframe to prevent errors
-# (Note: We don't filter COLS_GEN here strictly because we want to see if they are missing, 
-# but standard practice is to filter. Assuming populate.py logic fills missing ones with empty strings/NaN)
-COLS_LONG = [c for c in COLS_LONG if c in LEADERBOARD_DF.columns]
-
 # --- UI Layout ---
 demo = gr.Blocks(css=custom_css)
 
@@ -96,44 +95,58 @@ with demo:
             gr.Markdown("### Table 1: Comparison across LongText2Video, Entities2Video and Video2Video")
             
             # 1. Select Columns
-            # We use a list comprehension to ensure we only pick columns that actually exist in DF
-            # to avoid KeyErrors, but keep the order defined in COLS_GEN
             valid_gen_cols = [c for c in COLS_GEN if c in LEADERBOARD_DF.columns]
             df_gen = LEADERBOARD_DF[valid_gen_cols].copy()
             
-            # 2. Filter Rows (Only specific models)
-            target_models = ["UniVA", "LTX-Video", "Wan", "Seedance"]
-            # Ensure strict matching with 'Model' column
+            # 2. Filter Rows (Only specific models for Table 1)
+            target_models_gen = ["UniVA", "LTX-Video", "Wan", "Seedance"]
+            
             if "Model" in df_gen.columns:
-                df_gen = df_gen[df_gen["Model"].isin(target_models)]
-                
-                # 3. Sort rows by the order in target_models list
-                # This sets the order to: UniVA, LTX-Video, Wan, Seedance (or however you arranged target_models)
-                df_gen["Model"] = pd.Categorical(df_gen["Model"], categories=target_models, ordered=True)
+                df_gen = df_gen[df_gen["Model"].isin(target_models_gen)]
+                # Sort by defined order
+                df_gen["Model"] = pd.Categorical(df_gen["Model"], categories=target_models_gen, ordered=True)
                 df_gen = df_gen.sort_values("Model")
 
             gr.Dataframe(
                 value=df_gen,
                 headers=valid_gen_cols,
-                datatype="markdown", # Markdown allows links to render
+                datatype="markdown",
                 elem_id="leaderboard-table-gen",
                 interactive=False,
                 visible=True,
             )
 
-        # === Tab 2: Long-Video Tasks (Table 2) ===
-        with gr.TabItem("🧠 Long-Video Tasks", elem_id="tab-long", id=1):
-            gr.Markdown("### Long-Video Understanding, Editing & Segmentation")
+        # === Tab 2: Perception & Editing (Table 2) ===
+        with gr.TabItem("🧠 Perception & Editing", elem_id="tab-long", id=1):
+            gr.Markdown("### Table 2: Understanding, Editing, and Segmentation")
             
-            # Prepare data: Default sort by Understanding (Acc)
-            df_long = LEADERBOARD_DF[COLS_LONG]
-            sort_col = "LongVideo QA / Acc"
-            if sort_col in df_long.columns:
-                df_long = df_long.sort_values(by=sort_col, ascending=False)
+            # 1. Select Columns
+            valid_long_cols = [c for c in COLS_LONG if c in LEADERBOARD_DF.columns]
+            df_long = LEADERBOARD_DF[valid_long_cols].copy()
+
+            # 2. Filter Rows (Only specific models for Table 2)
+            # Removing pure generation models (Wan, LTX, Seedance) as they don't have these scores
+            target_models_long = [
+                "UniVA", 
+                "InternVL3-38B", 
+                "Qwen2.5-VL-72B", 
+                "Gemini 2.5 Pro", 
+                "GPT-4o", 
+                "Vace", 
+                "SA2VA"
+            ]
+
+            if "Model" in df_long.columns:
+                df_long = df_long[df_long["Model"].isin(target_models_long)]
+                
+                # 3. Sort: Default by Understanding Acc
+                sort_col = "LongVideo QA / Acc"
+                if sort_col in df_long.columns:
+                    df_long = df_long.sort_values(by=sort_col, ascending=False)
             
             gr.Dataframe(
                 value=df_long,
-                headers=COLS_LONG,
+                headers=valid_long_cols,
                 datatype="markdown",
                 elem_id="leaderboard-table-long",
                 interactive=False,