WIP

Author: yeshaokai

llava/action/benchmark.py

@@ -10,8 +10,8 @@
 n_frames = 4
 topk = 5
 action_representation = 'GT_random_narration'
-gpt_model = 'gpt-4o-mini-2024-07-18'
-#gpt_model = 'gpt-4o-2024-08-06'
+#gpt_model = 'gpt-4o-mini-2024-07-18'
+gpt_model = 'gpt-4o-2024-08-06'
 perspective = 'first_person'
 benchmark_testing = True
 

llava/action/chatgpt_utils.py

@@ -379,7 +379,6 @@ def __init__(self,
         if gen_type == 'avion' or gen_type == 'tim':                  
             self.mc_generator = ActionMultiChoiceGenerator(self.annotation_root)
             assert os.path.exists(self.prediction_file)
-            print ('prediction_file'*5, self.prediction_file)
             with open(self.prediction_file, 'r') as f:
                 self.action_model_predictions = json.load(f)
         else:

llava/action/make_visualizations.py

@@ -65,6 +65,26 @@ def visualize_with_gpt_with_tim(n_samples, question_type = 'mc_'):
     inferencer.multi_process_run(n_samples, disable_api_calling=False)    
 
 
+def visualize_with_gpt_with_avion(n_samples, question_type = 'mc_'):
+    """
+    Here we should test gpt-4o, gpt-4o-mini with different prompts
+    """
+    inferencer = GPTInferenceAnnotator(gpt_model,
+                                       root,
+                                       annotation_file,
+                                        gen_type = 'avion',
+                                        prediction_file = avion_prediction_file,
+                                        clip_length = n_frames,
+                                        question_type = question_type,
+                                        action_representation=action_representation,
+                                        perspective = perspective,
+                                        benchmark_testing = benchmark_testing,
+                                        do_visualization = True,
+                                        topk = topk) 
+    
+    inferencer.multi_process_run(n_samples, disable_api_calling=False) 
+
 if __name__ == '__main__':
     #visualize_with_random(1, question_type = "mc_")
-    visualize_with_gpt_with_tim(1, question_type = "mc_")
+    #visualize_with_gpt_with_tim(1, question_type = "mc_")
+    visualize_with_gpt_with_avion(1, question_type = "mc_")