vbench2 filter

app.py

@@ -18,6 +18,12 @@ HF_TOKEN = os.environ.get("HF_TOKEN")
 global data_component, filter_component
 
 
+category_to_dimension = {}
+for key, value in VBENCH2_DIM2CAT.items():
+    if value not in category_to_dimension:
+        category_to_dimension[value] = []
+    category_to_dimension[value].append(key)
+
 def upload_file(files):
     file_paths = [file.name for file in files]
     return file_paths
@@ -384,12 +390,6 @@ def get_final_score_quality(df, selected_columns):
     return df
 
 def get_final_score2(df, selected_columns):
-    category_to_dimension = {}
-
-    for key, value in VBENCH2_DIM2CAT.items():
-        if value not in category_to_dimension:
-            category_to_dimension[value] = []
-        category_to_dimension[value].append(key)
     score_names = []
     for cur_score in category_to_dimension:
         score_name = f"{cur_score} Score"
@@ -497,11 +497,11 @@ def get_all_df_long(selected_columns, dir=LONG_DIR):
     df = df.sort_values(by="Selected Score", ascending=False)
     return df
 
-def get_all_df2(dir=VBENCH2_DIR):
+def get_all_df2(selected_columns, dir=VBENCH2_DIR):
     submission_repo = Repository(local_dir=SUBMISSION_NAME, clone_from=SUBMISSION_URL, use_auth_token=HF_TOKEN, repo_type="dataset")
     submission_repo.git_pull()
     df = pd.read_csv(dir)
-    df = get_final_score2(df, TASK_INFO_2)
+    df = get_final_score2(df, selected_columns)
     df = df.sort_values(by="Total Score", ascending=False)
     return df
     
@@ -621,13 +621,13 @@ def on_filter_model_size_method_change_long(selected_columns, vbench_team_sample
         visible=True,
         )
     return filter_component#.value
+ 
 
-
-def on_filter_model_size_method_change_2(vbench_team_sample, vbench_team_eval=False):
-    updated_data = get_all_df2(VBENCH2_DIR)
+def on_filter_model_size_method_change_2(selected_columns=TASK_INFO_2, vbench_team_sample=False, vbench_team_eval=False):
+    updated_data = get_all_df2(selected_columns, VBENCH2_DIR)
     print(updated_data)
     if vbench_team_sample:
-        updated_data = updated_data[updated_data["Sampled by"] == 'VBench Team'] 
+        updated_data = updated_data[updated_data["Sampled by"] == 'VBench Team']
     if vbench_team_eval:
         updated_data = updated_data[updated_data['Evaluated by'] == 'VBench Team']
    
@@ -642,7 +642,34 @@ def on_filter_model_size_method_change_2(vbench_team_sample, vbench_team_eval=Fa
         interactive=False,
         visible=True,
         )
-    return filter_component#.value
+    return filter_component
+
+def on_filter_model_size_method_score_change_2(select_score, vbench_team_sample=False, vbench_team_eval=False):
+    selected_columns = category_to_dimension[select_score]
+    print(select_score,"===>",selected_columns)
+    updated_data = get_all_df2(selected_columns, VBENCH2_DIR)
+    print(updated_data)
+    if vbench_team_sample:
+        updated_data = updated_data[updated_data["Sampled by"] == 'VBench Team']
+    if vbench_team_eval:
+        updated_data = updated_data[updated_data['Evaluated by'] == 'VBench Team']
+   
+    present_columns = VBENCH2_MODEL_INFO_DEFAULT + [f"{select_score} Score"] + selected_columns
+    updated_headers = present_columns
+    updated_data = updated_data[present_columns]
+    updated_data = updated_data.sort_values(by=f"{select_score} Score", ascending=False)
+    updated_data = convert_scores_to_percentage(updated_data)
+    update_datatype = [VBENCH2_TITLE_TYPE[COLUMN_NAMES_2.index(x)] for x in updated_headers]
+    print(updated_data)
+    filter_component = gr.components.Dataframe(
+        value=updated_data, 
+        headers=updated_headers,
+        type="pandas", 
+        datatype=update_datatype,
+        interactive=False,
+        visible=True,
+        )
+    return filter_component, gr.update(value=selected_columns)
 
 block = gr.Blocks()
 
@@ -699,7 +726,7 @@ with block:
                 datatype=DATA_TITILE_TYPE,
                 interactive=False,
                 visible=True,
-                height=700,
+                # height=700,
                 )
     
             choosen_q.click(choose_all_quailty, inputs=None, outputs=[checkbox_group]).then(fn=on_filter_model_size_method_change, inputs=[ checkbox_group, vbench_team_filter,vbench_validate_filter], outputs=data_component)
@@ -724,18 +751,29 @@ with block:
                 TABLE_INTRODUCTION
             )
             with gr.Row():
-                with gr.Column():
-                    vbench_team_filter_2 = gr.Checkbox(
-                            label="Sampled by VBench Team (Uncheck to view all submissions)",
-                            value=False,
-                            interactive=True
-                        )
-                    vbench_validate_filter_2 = gr.Checkbox(
-                        label="Evaluated by VBench Team (Uncheck to view all submissions)",
-                        value=True,
+                vbench_team_filter_2 = gr.Checkbox(
+                        label="Sampled by VBench Team (Uncheck to view all submissions)",
+                        value=False,
                         interactive=True
                     )
-
+                vbench_validate_filter_2 = gr.Checkbox(
+                    label="Evaluated by VBench Team (Uncheck to view all submissions)",
+                    value=True,
+                    interactive=True
+                )
+            with gr.Row():
+                vbench2_creativity_button = gr.Button("Show Creativity Score")
+                vbench2_commonsense_button = gr.Button("Show Commonsense Score")
+                vbench2_control_button = gr.Button("Show Controllability Score")
+                vbench2_human_button = gr.Button("Show Human Fidelity Score")
+                vbench2_physics_button = gr.Button("Show Physics Score")
+            with gr.Row():
+                vbench2_checkgroup = gr.CheckboxGroup(
+                        choices=TASK_INFO_2,
+                        value=TASK_INFO_2,
+                        label="Evaluation Dimension",
+                        interactive=True,
+                    )
             
             data_component_2 = gr.components.Dataframe(
                 value=get_baseline_df_2, 
@@ -744,11 +782,17 @@ with block:
                 datatype=VBENCH2_TITLE_TYPE,
                 interactive=False,
                 visible=True,
-                height=700,
+                # height=700,
                 )
-            vbench_team_filter_2.change(fn=on_filter_model_size_method_change_2, inputs=[vbench_team_filter_2, vbench_validate_filter], outputs=data_component_2)
-            vbench_validate_filter_2.change(fn=on_filter_model_size_method_change_2, inputs=[vbench_team_filter_2, vbench_validate_filter], outputs=data_component_2)
-        
+            vbench_team_filter_2.change(fn=on_filter_model_size_method_change_2, inputs=[vbench2_checkgroup, vbench_team_filter_2, vbench_validate_filter], outputs=data_component_2)
+            vbench_validate_filter_2.change(fn=on_filter_model_size_method_change_2, inputs=[vbench2_checkgroup, vbench_team_filter_2, vbench_validate_filter], outputs=data_component_2)
+            # vbench2_checkgroup.change(fn=on_filter_model_size_method_change_2, inputs=[vbench2_checkgroup, vbench_team_filter_2, vbench_validate_filter], outputs=data_component_2)
+            vbench2_creativity_button.click(fn=on_filter_model_size_method_score_change_2, inputs=[gr.State("Creativity"), vbench_team_filter_2, vbench_validate_filter], outputs=[data_component_2, vbench2_checkgroup])
+            vbench2_commonsense_button.click(fn=on_filter_model_size_method_score_change_2, inputs=[gr.State("Commonsense"), vbench_team_filter_2, vbench_validate_filter], outputs=[data_component_2, vbench2_checkgroup])
+            vbench2_control_button.click(fn=on_filter_model_size_method_score_change_2, inputs=[gr.State("Controllability"), vbench_team_filter_2, vbench_validate_filter], outputs=[data_component_2, vbench2_checkgroup])
+            vbench2_human_button.click(fn=on_filter_model_size_method_score_change_2, inputs=[gr.State("Human Fidelity"), vbench_team_filter_2, vbench_validate_filter], outputs=[data_component_2, vbench2_checkgroup])
+            vbench2_physics_button.click(fn=on_filter_model_size_method_score_change_2, inputs=[gr.State("Physics"), vbench_team_filter_2, vbench_validate_filter], outputs=[data_component_2, vbench2_checkgroup])
+
         with gr.TabItem("Video Quality", elem_id="vbench-tab-table", id=3):
             with gr.Accordion("INSTRUCTION", open=False):
                     citation_button = gr.Textbox(
@@ -868,7 +912,7 @@ with block:
                 datatype=DATA_TITILE_TYPE,
                 interactive=False,
                 visible=True,
-                height=700,
+                # height=700,
                 )
     
             choosen_q_long.click(choose_all_quailty, inputs=None, outputs=[checkbox_group_long]).then(fn=on_filter_model_size_method_change_long, inputs=[ checkbox_group_long, vbench_team_filter_long, vbench_validate_filter_long], outputs=data_component)

constants.py

@@ -50,13 +50,16 @@ TASK_INFO = [
 ]
 
 
-MODEL_INFO_2 = [
+VBENCH2_MODEL_INFO_DEFAULT = [
     "Model (alphabetical order)",
     "Sampled by",
     "Evaluated by",
     "Accessibility",
     "Date",
-    "Total Score",
+    "Total Score"
+]
+
+VBENCH2_DIM_DEFAULT_INFO = [
     'Creativity Score', 
     'Commonsense Score', 
     'Controllability Score', 
@@ -64,6 +67,8 @@ MODEL_INFO_2 = [
     'Physics Score'
 ]
 
+MODEL_INFO_2 = VBENCH2_MODEL_INFO_DEFAULT + VBENCH2_DIM_DEFAULT_INFO
+
 TASK_INFO_2 = [
     "Human Anatomy",
     "Human Clothes",