update

app.py

@@ -3,24 +3,41 @@ import pandas as pd
 from utils import load_all_results
 
 
-def sort_leaderboard(sort_by="Video-MME", descending=True):
+def filter_and_sort_leaderboard(category_filters, sort_by="Video-MME", descending=True):
     df = load_all_results()
-    return df.sort_values(by=sort_by, ascending=not descending).reset_index(drop=True)
+    filtered_df = df[df["Category"].isin(category_filters)]
+    return filtered_df.sort_values(by=sort_by, ascending=not descending).reset_index(drop=True)
 
 
 def build_interface():
     with gr.Blocks() as demo:
-        gr.Markdown("## 📊 LVU VLM Leaderboard")
-        gr.Markdown("Benchmarks: Video-MME, MLVU, LVBench, LongVideoBench")
+        gr.Markdown("## 🎥 LVU VLM Leaderboard")
+        gr.Markdown("Benchmarks: Video-MME, MLVU, LongVideoBench")
 
         with gr.Row():
-            sort_metric = gr.Dropdown(choices=["Video-MME", "MLVU", "LVBench", "LongVideoBench"], label="Sort by", value="Video-MME")
+            sort_metric = gr.Dropdown(choices=["Video-MME", "MLVU", "LongVideoBench"], label="Sort by", value="Video-MME")
             descending = gr.Checkbox(label="Descending", value=True)
 
+        with gr.Row():
+            proprietary = gr.Checkbox(label="Proprietary Models", value=True)
+            transformer = gr.Checkbox(label="Open-source Transformer-based LMMs", value=True)
+            efficient = gr.Checkbox(label="Open-source Efficient LMMs", value=True)
+
         leaderboard = gr.Dataframe(label="Leaderboard")
 
-        sort_metric.change(fn=sort_leaderboard, inputs=[sort_metric, descending], outputs=leaderboard)
-        descending.change(fn=sort_leaderboard, inputs=[sort_metric, descending], outputs=leaderboard)
+        def update_board(sort_by, descending, proprietary, transformer, efficient):
+            filters = []
+            if proprietary:
+                filters.append("Proprietary Models")
+            if transformer:
+                filters.append("Open-source Transformer-based LMMs")
+            if efficient:
+                filters.append("Open-source Efficient LMMs")
+            return filter_and_sort_leaderboard(filters, sort_by, descending)
+
+        inputs = [sort_metric, descending, proprietary, transformer, efficient]
+        for inp in inputs:
+            inp.change(fn=update_board, inputs=inputs, outputs=leaderboard)
 
     return demo
 

constants.py

@@ -4,3 +4,26 @@ HF_TOKEN = os.environ.get("HF_TOKEN")
 
 
 RESULTS_DIR = "static/eval_results"
+
+
+CATEGORY_MAP = {
+    "GPT-4V": ("-", "Proprietary Models"),
+    "GPT-4o": ("-", "Proprietary Models"),
+    "Gemini-1.5-Pro": ("-", "Proprietary Models"),
+
+    "VideoChat2": ("7B", "Open-source Transformer-based LMMs"),
+    "ShareGPT4Video": ("7B", "Open-source Transformer-based LMMs"),
+    "LongVA": ("7B", "Open-source Transformer-based LMMs"),
+    "Video-CCAM": ("9B", "Open-source Transformer-based LMMs"),
+    "Kangaroo": ("8B", "Open-source Transformer-based LMMs"),
+    "InternVL2": ("8B", "Open-source Transformer-based LMMs"),
+    "LLaVA-OneVision": ("7B", "Open-source Transformer-based LMMs"),
+    "Qwen2-VL": ("7B", "Open-source Transformer-based LMMs"),
+    "Phi-4-Mini": ("5.6B", "Open-source Transformer-based LMMs"),
+
+    "LLaVA-Mini": ("7B", "Open-source Efficient LMMs"),
+    "LongLLaVA": ("9B", "Open-source Efficient LMMs"),
+    "LongVU": ("9B", "Open-source Efficient LMMs"),
+    "Video-XL": ("7B", "Open-source Efficient LMMs"),
+    "VAMBA": ("10B", "Open-source Efficient LMMs")
+}

utils.py

@@ -1,7 +1,7 @@
 import os
 import json
 import pandas as pd
-from constants import RESULTS_DIR
+from constants import RESULTS_DIR, CATEGORY_MAP
 
 
 def load_all_results():
@@ -10,8 +10,10 @@ def load_all_results():
         if file.endswith(".json"):
             with open(os.path.join(RESULTS_DIR, file), "r") as f:
                 data = json.load(f)
-                if isinstance(data, list):
-                    all_data.extend(data)
-                elif isinstance(data, dict):
-                    all_data.append(data)
+                for entry in data:
+                    model = entry["Model"]
+                    size, category = CATEGORY_MAP.get(model, ("N/A", "Unknown"))
+                    entry["Size"] = size
+                    entry["Category"] = category
+                    all_data.append(entry)
     return pd.DataFrame(all_data)