Update app.py

app.py

@@ -5,7 +5,6 @@ import torch
 from decord import cpu, VideoReader, bridge
 from transformers import AutoModelForCausalLM, AutoTokenizer
 from transformers import BitsAndBytesConfig
-import json
 
 # Model Configuration
 MODEL_PATH = "THUDM/cogvlm2-llama3-caption"
@@ -100,20 +99,26 @@ def predict(prompt, video_data, temperature, model, tokenizer):
     return response
 
 def get_analysis_prompt(step_number, possible_reasons):
-    return f"""Analyze the video of Step {step_number} in the tire manufacturing process.
-    
-Possible delay reasons for this step are:
+    return f"""You are an AI expert system specialized in analyzing manufacturing processes and identifying production delays in tire manufacturing. Your role is to accurately classify delay reasons based on visual evidence from production line footage.
+
+Task Context:
+You are analyzing video footage from Step {step_number} of a tire manufacturing process where a delay has been detected. Your task is to determine the most likely cause of the delay from the following possible reasons:
 {', '.join(possible_reasons)}
 
-Based on the video evidence, determine which of these reasons best explains the delay.
-Please provide:
-1. Your chosen reason from the list above
-2. Specific visual evidence supporting this choice
-3. Brief explanation of why other reasons are less likely
+Required Analysis:
+1. Carefully observe the video for visual cues indicating production interruption
+2. Compare observed evidence against each possible delay reason
+3. Select the most likely reason based on visual evidence
 
-Focus your analysis on visual cues that support your conclusion."""
+Please provide your analysis in the following format:
+1. Selected Reason: [State the most likely reason from the given options]
+2. Visual Evidence: [Describe specific visual cues that support your selection]
+3. Reasoning: [Explain why this reason best matches the observed evidence]
+4. Alternative Analysis: [Brief explanation of why other possible reasons are less likely]
 
-def inference(video, step_number, selected_reason):
+Important: Base your analysis solely on visual evidence from the video. Focus on concrete, observable details rather than assumptions."""
+
+def inference(video, step_number):
     if not video:
         return "Please upload a video first."
     
@@ -136,13 +141,15 @@ def inference(video, step_number, selected_reason):
     except Exception as e:
         return f"An error occurred: {str(e)}"
 
-def update_reasons(step):
-    """Update the dropdown choices based on the selected step"""
-    return gr.Dropdown(choices=DELAY_REASONS[step])
-
 # Gradio Interface
 def create_interface():
     with gr.Blocks() as demo:
+        gr.Markdown("""
+        # Manufacturing Delay Analysis System
+        Upload a video of the manufacturing step and select the step number. 
+        The system will analyze the video and determine the most likely cause of delay.
+        """)
+        
         with gr.Row():
             with gr.Column():
                 video = gr.Video(label="Upload Manufacturing Video", sources=["upload"])
@@ -151,27 +158,15 @@ def create_interface():
                     label="Manufacturing Step",
                     value="Step 1"
                 )
-                reason = gr.Dropdown(
-                    choices=DELAY_REASONS["Step 1"],
-                    label="Select Delay Reason",
-                    value=DELAY_REASONS["Step 1"][0]
-                )
                 analyze_btn = gr.Button("Analyze Delay", variant="primary")
             
             with gr.Column():
                 output = gr.Textbox(label="Analysis Result", lines=10)
         
-        # Update reasons when step changes
-        step_number.change(
-            fn=update_reasons,
-            inputs=[step_number],
-            outputs=[reason]
-        )
-        
         # Trigger analysis when button is clicked
         analyze_btn.click(
             fn=inference,
-            inputs=[video, step_number, reason],
+            inputs=[video, step_number],
             outputs=[output]
         )