Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -139,16 +139,14 @@ CHUNK_FRAME_COUNT = int(VIDEO_FPS * CHUNK_SIZE_SECONDS)
|
|
139 |
TEMP_VIDEO_FILE = "temp_video_chunk.mp4"
|
140 |
|
141 |
def process_video_chunk(frame, state_dict):
|
142 |
-
"""
|
143 |
-
Processes a single frame, adds it to a buffer, and encodes a video chunk
|
144 |
-
when the buffer is full. Now includes detailed timing logs.
|
145 |
-
"""
|
146 |
if frame is None:
|
147 |
return None, "Status: Inactive", None, state_dict
|
148 |
|
149 |
# Unpack the state
|
150 |
frame_buffer = state_dict['buffer']
|
151 |
start_time = state_dict['start_time']
|
|
|
|
|
152 |
|
153 |
# If the buffer is empty, this is the first frame of a new chunk. Record start time.
|
154 |
if not frame_buffer:
|
@@ -165,45 +163,52 @@ def process_video_chunk(frame, state_dict):
|
|
165 |
level = indic.get("drowsiness_level", "Awake")
|
166 |
lighting = indic.get("lighting", "Good")
|
167 |
score = indic.get("details", {}).get("Score", 0.0)
|
168 |
-
# The status text is updated on every single frame.
|
169 |
-
status_txt = f"Lighting: {lighting}\nStatus: {level}\nScore: {score:.2f}"
|
170 |
-
|
171 |
audio_payload = alert_manager.trigger_alert(level, lighting)
|
172 |
audio_out = gr.Audio(value=audio_payload, autoplay=True) if audio_payload else None
|
173 |
|
174 |
# --- Video Buffering Logic ---
|
175 |
frame_buffer.append(processed_frame)
|
176 |
|
177 |
-
video_out = None
|
178 |
if len(frame_buffer) >= CHUNK_FRAME_COUNT:
|
179 |
-
# --- NEW: Logging buffer fill time ---
|
180 |
buffer_fill_time = time.perf_counter() - start_time
|
181 |
logging.info(f"BUFFER: Filled {len(frame_buffer)} frames in {buffer_fill_time:.2f} seconds.")
|
182 |
-
|
183 |
-
# --- NEW: Logging encoding time ---
|
184 |
-
encoding_start_time = time.perf_counter()
|
185 |
|
186 |
-
|
187 |
h, w, _ = frame_buffer[0].shape
|
188 |
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
|
189 |
writer = cv2.VideoWriter(TEMP_VIDEO_FILE, fourcc, VIDEO_FPS, (w, h))
|
190 |
for f in frame_buffer:
|
191 |
writer.write(f)
|
192 |
writer.release()
|
193 |
-
|
194 |
encoding_time = time.perf_counter() - encoding_start_time
|
195 |
logging.info(f"ENCODING: Video chunk encoded in {encoding_time:.2f} seconds.")
|
|
|
|
|
|
|
|
|
|
|
|
|
196 |
|
197 |
-
|
198 |
-
|
199 |
-
|
|
|
|
|
|
|
|
|
200 |
|
201 |
# Pack state back up to return
|
202 |
-
new_state = {
|
|
|
|
|
|
|
|
|
|
|
203 |
|
204 |
-
# Note: Status and Audio are returned on every frame for real-time feedback
|
205 |
return video_out, status_txt, audio_out, new_state
|
206 |
|
|
|
207 |
# βββββββββββββββββββββββββββββ UI Definition
|
208 |
def create_readme_tab():
|
209 |
"""Creates the content for the 'About' tab."""
|
@@ -284,7 +289,7 @@ def create_detection_tab():
|
|
284 |
)
|
285 |
|
286 |
def create_video_experiment_tab():
|
287 |
-
"""Creates the content for the Video Chunk experiment tab."""
|
288 |
with gr.Blocks() as video_tab:
|
289 |
gr.Markdown("## π§ͺ Video Output Experiment")
|
290 |
gr.Markdown(f"This feed buffers processed frames and outputs them as **{CHUNK_SIZE_SECONDS}-second video chunks**. Notice the trade-off between smoothness and latency. Alerts remain real-time.")
|
@@ -292,20 +297,29 @@ def create_video_experiment_tab():
|
|
292 |
with gr.Column(scale=2):
|
293 |
cam_video = gr.Image(sources=["webcam"], streaming=True, label="Live Camera Feed")
|
294 |
with gr.Column(scale=1):
|
295 |
-
|
296 |
-
|
|
|
|
|
297 |
out_audio_video = gr.Audio(label="Alert", autoplay=True, visible=False)
|
298 |
|
299 |
-
# NEW: State is
|
300 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
301 |
|
302 |
cam_video.stream(
|
303 |
fn=process_video_chunk,
|
304 |
-
inputs=[cam_video,
|
305 |
-
outputs=[out_video, out_text_video, out_audio_video,
|
306 |
)
|
307 |
return video_tab
|
308 |
-
|
|
|
309 |
with gr.Blocks(title="π Drive Paddy β Drowsiness Detection", theme=gr.themes.Soft()) as app:
|
310 |
gr.Markdown("# π **Drive Paddy**")
|
311 |
with gr.Tabs():
|
|
|
139 |
TEMP_VIDEO_FILE = "temp_video_chunk.mp4"
|
140 |
|
141 |
def process_video_chunk(frame, state_dict):
|
|
|
|
|
|
|
|
|
142 |
if frame is None:
|
143 |
return None, "Status: Inactive", None, state_dict
|
144 |
|
145 |
# Unpack the state
|
146 |
frame_buffer = state_dict['buffer']
|
147 |
start_time = state_dict['start_time']
|
148 |
+
last_buffer_fill_time = state_dict['last_buffer_fill_time']
|
149 |
+
last_encoding_time = state_dict['last_encoding_time']
|
150 |
|
151 |
# If the buffer is empty, this is the first frame of a new chunk. Record start time.
|
152 |
if not frame_buffer:
|
|
|
163 |
level = indic.get("drowsiness_level", "Awake")
|
164 |
lighting = indic.get("lighting", "Good")
|
165 |
score = indic.get("details", {}).get("Score", 0.0)
|
|
|
|
|
|
|
166 |
audio_payload = alert_manager.trigger_alert(level, lighting)
|
167 |
audio_out = gr.Audio(value=audio_payload, autoplay=True) if audio_payload else None
|
168 |
|
169 |
# --- Video Buffering Logic ---
|
170 |
frame_buffer.append(processed_frame)
|
171 |
|
172 |
+
video_out = None
|
173 |
if len(frame_buffer) >= CHUNK_FRAME_COUNT:
|
|
|
174 |
buffer_fill_time = time.perf_counter() - start_time
|
175 |
logging.info(f"BUFFER: Filled {len(frame_buffer)} frames in {buffer_fill_time:.2f} seconds.")
|
|
|
|
|
|
|
176 |
|
177 |
+
encoding_start_time = time.perf_counter()
|
178 |
h, w, _ = frame_buffer[0].shape
|
179 |
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
|
180 |
writer = cv2.VideoWriter(TEMP_VIDEO_FILE, fourcc, VIDEO_FPS, (w, h))
|
181 |
for f in frame_buffer:
|
182 |
writer.write(f)
|
183 |
writer.release()
|
|
|
184 |
encoding_time = time.perf_counter() - encoding_start_time
|
185 |
logging.info(f"ENCODING: Video chunk encoded in {encoding_time:.2f} seconds.")
|
186 |
+
|
187 |
+
video_out = TEMP_VIDEO_FILE
|
188 |
+
last_buffer_fill_time = buffer_fill_time
|
189 |
+
last_encoding_time = encoding_time
|
190 |
+
frame_buffer = []
|
191 |
+
start_time = None
|
192 |
|
193 |
+
# --- NEW: Update status text to include buffer/encoding times ---
|
194 |
+
status_txt = (
|
195 |
+
f"Status: {level} (Score: {score:.2f})\n"
|
196 |
+
f"Buffer Fill Time: {last_buffer_fill_time:.2f}s\n"
|
197 |
+
f"Video Encode Time: {last_encoding_time:.2f}s"
|
198 |
+
)
|
199 |
+
logging.info(f"Status: {status_txt}")
|
200 |
|
201 |
# Pack state back up to return
|
202 |
+
new_state = {
|
203 |
+
'buffer': frame_buffer,
|
204 |
+
'start_time': start_time,
|
205 |
+
'last_buffer_fill_time': last_buffer_fill_time,
|
206 |
+
'last_encoding_time': last_encoding_time
|
207 |
+
}
|
208 |
|
|
|
209 |
return video_out, status_txt, audio_out, new_state
|
210 |
|
211 |
+
|
212 |
# βββββββββββββββββββββββββββββ UI Definition
|
213 |
def create_readme_tab():
|
214 |
"""Creates the content for the 'About' tab."""
|
|
|
289 |
)
|
290 |
|
291 |
def create_video_experiment_tab():
|
292 |
+
"""Creates the content for the Video Chunk experiment tab - UPDATED."""
|
293 |
with gr.Blocks() as video_tab:
|
294 |
gr.Markdown("## π§ͺ Video Output Experiment")
|
295 |
gr.Markdown(f"This feed buffers processed frames and outputs them as **{CHUNK_SIZE_SECONDS}-second video chunks**. Notice the trade-off between smoothness and latency. Alerts remain real-time.")
|
|
|
297 |
with gr.Column(scale=2):
|
298 |
cam_video = gr.Image(sources=["webcam"], streaming=True, label="Live Camera Feed")
|
299 |
with gr.Column(scale=1):
|
300 |
+
# --- NEW: Use the blank video as the initial value ---
|
301 |
+
out_video = gr.Video(label="Processed Video Chunk", value=BLANK_VIDEO_FILE)
|
302 |
+
# --- NEW: Textbox is now larger to show more info ---
|
303 |
+
out_text_video = gr.Textbox(label="Live Status", lines=4, interactive=False)
|
304 |
out_audio_video = gr.Audio(label="Alert", autoplay=True, visible=False)
|
305 |
|
306 |
+
# --- NEW: State is a dictionary to hold the buffer and timing info ---
|
307 |
+
initial_state = {
|
308 |
+
'buffer': [],
|
309 |
+
'start_time': None,
|
310 |
+
'last_buffer_fill_time': 0.0,
|
311 |
+
'last_encoding_time': 0.0
|
312 |
+
}
|
313 |
+
state = gr.State(initial_state)
|
314 |
|
315 |
cam_video.stream(
|
316 |
fn=process_video_chunk,
|
317 |
+
inputs=[cam_video, state],
|
318 |
+
outputs=[out_video, out_text_video, out_audio_video, state]
|
319 |
)
|
320 |
return video_tab
|
321 |
+
|
322 |
+
|
323 |
with gr.Blocks(title="π Drive Paddy β Drowsiness Detection", theme=gr.themes.Soft()) as app:
|
324 |
gr.Markdown("# π **Drive Paddy**")
|
325 |
with gr.Tabs():
|