Update app.py

app.py

@@ -19,7 +19,7 @@ api_key = os.getenv("OPEN_AI_KEY")
 user_name = os.getenv("USER_NAME")
 password = os.getenv("PASSWORD")
 
-LENGTH = 3
+LENGTH = 3  # Duration of each chunk in seconds
 WEBCAM = 0
 
 MARKDOWN = """
@@ -32,7 +32,7 @@ AVATARS = (
 
 # Set your OpenAI API key
 openai.api_key = api_key
-MODEL="gpt-4o"
+MODEL = "gpt-4o"
 client = openai.OpenAI(api_key=api_key)
 
 # Global variable to stop the video capture loop
@@ -69,7 +69,6 @@ def clip_video_segment(input_video_path, start_time, duration):
     return output_video_path
 
 def encode_to_video_fast(frames, fps):
-    
     os.makedirs('videos', exist_ok=True)
     video_clip_path = f"videos/{uuid.uuid4()}.mp4"
 
@@ -88,7 +87,6 @@ def encode_to_video_fast(frames, fps):
 
     return video_clip_path
 
-
 def encode_to_video(frames, fps):
     os.makedirs('videos', exist_ok=True)
     video_clip_path = f"videos/{uuid.uuid4()}.mp4"
@@ -104,9 +102,9 @@ def encode_to_video(frames, fps):
     return video_clip_path
 
 # Function to process video frames using GPT-4 API
-def process_frames(frames, frames_to_skip = 1):
+def process_frames(frames, frames_to_skip=1):
     os.makedirs('saved_frames', exist_ok=True)
-    curr_frame=0
+    curr_frame = 0
     base64Frames = []
     while curr_frame < len(frames) - 1:
         _, buffer = cv2.imencode(".jpg", frames[curr_frame])
@@ -119,8 +117,6 @@ def check_condition(prompt, base64Frames):
     start_time = time.time()
     print('checking condition for frames:', len(base64Frames))
 
-        # Save frames as images
-
     try:
         messages = [
             {"role": "system", "content": """You are analyzing video to check if the user's condition is met. 
@@ -133,7 +129,7 @@ def check_condition(prompt, base64Frames):
             model="gpt-4o",
             messages=messages,
             temperature=0,
-            response_format={ "type": "json_object" }
+            response_format={"type": "json_object"}
         )
     
         end_time = time.time()
@@ -151,7 +147,6 @@ def check_condition(prompt, base64Frames):
     except:
         print('result', response.usage.total_tokens, api_response)
         return frames_count, processing_time, api_response
-    
 
 # Function to process video clip and update the chatbot
 def process_clip(prompt, frames, chatbot):
@@ -183,19 +178,23 @@ def process_clip_from_file(prompt, frames, chatbot, fps, video_path, id):
     global stop_capture
     if not stop_capture:
         israel_tz = pytz.timezone('Asia/Jerusalem')
-        start_time = datetime.now(israel_tz).strftime('%H:%M:%S')
-        print("[Start]:", start_time, len(frames))
+        start_time_str = datetime.now(israel_tz).strftime('%H:%M:%S')
+        print("[Start]:", start_time_str, len(frames))
         
         frames_to_skip = int(fps)
         base64Frames = process_frames(frames, frames_to_skip)
         frames_count, processing_time, api_response = check_condition(prompt, base64Frames)
         
-        result = None
         if api_response and api_response.get("condition_met", False):
-            # video_clip_path = encode_to_video_fast(frames, fps)
-            video_clip_path = clip_video_segment_2(video_path, id*LENGTH, LENGTH)
-            chatbot.append(((video_clip_path,), None))
-            chatbot.append((f"Event ID: {id+1}\nDetails: {api_response.get('details', '')}", None))
+            # Calculate start and end times
+            start_time = id * LENGTH
+            end_time = start_time + LENGTH
+
+            # Format times
+            start_time_formatted = time.strftime('%H:%M:%S', time.gmtime(start_time))
+            end_time_formatted = time.strftime('%H:%M:%S', time.gmtime(end_time))
+
+            chatbot.append((f"Event ID: {id+1}\nStart Time: {start_time_formatted}\nEnd Time: {end_time_formatted}\nDetails: {api_response.get('details', '')}", None))
     
     return chatbot
 
@@ -203,20 +202,23 @@ def process_clip_from_file(prompt, frames, chatbot, fps, video_path, id):
 def process_clip_from_file_sync(prompt, frames, fps, video_path, id):
     global stop_capture
     if not stop_capture:
-        israel_tz = pytz.timezone('Asia/Jerusalem')
-        start_time = datetime.now(israel_tz).strftime('%H:%M:%S')
-        print("[Start]:", start_time, len(frames))
-        
+        # Calculate start and end times in seconds
+        start_time = id * LENGTH
+        end_time = start_time + LENGTH
+
+        # Format times as HH:MM:SS
+        start_time_formatted = time.strftime('%H:%M:%S', time.gmtime(start_time))
+        end_time_formatted = time.strftime('%H:%M:%S', time.gmtime(end_time))
+
         frames_to_skip = int(fps)
         base64Frames = process_frames(frames, frames_to_skip)
         frames_count, processing_time, api_response = check_condition(prompt, base64Frames)
         
         if api_response and api_response.get("condition_met", False):
-            video_clip_path = clip_video_segment_2(video_path, id*LENGTH, LENGTH)
             event = {
                 'event_id': id + 1,
-                'video_clip_path': video_clip_path,
-                'start_time': start_time,
+                'start_time': start_time_formatted,
+                'end_time': end_time_formatted,
                 'details': api_response.get('details', '')
             }
             return event
@@ -227,7 +229,6 @@ def analyze_stream(prompt, stream, chatbot):
     global stop_capture
     stop_capture = False
 
-
     cap = cv2.VideoCapture(stream or WEBCAM)
 
     frames = []
@@ -257,7 +258,7 @@ def analyze_video_file(prompt, video_path, chatbot):
     
     # Get video properties
     fps = int(cap.get(cv2.CAP_PROP_FPS))  # Frames per second
-    frames_per_chunk = fps * LENGTH  # Number of frames per LENGTH-second chunk
+    frames_per_chunk = int(fps * LENGTH)  # Number of frames per LENGTH-second chunk
     
     frames = []
     chunk = 0
@@ -276,12 +277,12 @@ def analyze_video_file(prompt, video_path, chatbot):
             if len(frames) >= frames_per_chunk:
                 futures.append(executor.submit(process_clip_from_file, prompt, frames.copy(), chatbot, fps, video_path, chunk))
                 frames = []
-                chunk+=1
+                chunk += 1
         
         # If any remaining frames that are less than LENGTH seconds, process them as a final chunk
         if len(frames) > 0:
             futures.append(executor.submit(process_clip_from_file, prompt, frames.copy(), chatbot, fps, video_path, chunk))
-            chunk+=1
+            chunk += 1
         
         cap.release()
         # Yield results as soon as each thread completes
@@ -299,7 +300,7 @@ def analyze_video_file_sync(prompt, video_path):
     
     # Get video properties
     fps = int(cap.get(cv2.CAP_PROP_FPS))  # Frames per second
-    frames_per_chunk = fps * LENGTH  # Number of frames per LENGTH-second chunk
+    frames_per_chunk = int(fps * LENGTH)  # Number of frames per LENGTH-second chunk
     
     frames = []
     chunk = 0
@@ -319,12 +320,12 @@ def analyze_video_file_sync(prompt, video_path):
             if len(frames) >= frames_per_chunk:
                 futures.append(executor.submit(process_clip_from_file_sync, prompt, frames.copy(), fps, video_path, chunk))
                 frames = []
-                chunk+=1
+                chunk += 1
         
         # If any remaining frames that are less than LENGTH seconds, process them as a final chunk
         if len(frames) > 0:
             futures.append(executor.submit(process_clip_from_file_sync, prompt, frames.copy(), fps, video_path, chunk))
-            chunk+=1
+            chunk += 1
         
         cap.release()
         # Collect results as threads complete
@@ -364,7 +365,7 @@ with gr.Blocks(title="Conntour", fill_height=True) as demo:
     # Add new API endpoint (without UI components)
     with gr.Row(visible=False) as hidden_api:
         api_prompt = gr.Textbox(label="Prompt")
-        api_video = gr.Textbox(label="Prompt")
+        api_video = gr.Video(label="Video File")
         api_output = gr.JSON(label="Captured Events")
         api_btn = gr.Button("Analyze Video File")