Update visualization.py
Browse files- visualization.py +60 -0
visualization.py
CHANGED
@@ -199,3 +199,63 @@ def plot_posture(df, posture_scores, color='blue', anomaly_threshold=3):
|
|
199 |
plt.tight_layout()
|
200 |
plt.close()
|
201 |
return fig
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
199 |
plt.tight_layout()
|
200 |
plt.close()
|
201 |
return fig
|
202 |
+
|
203 |
+
|
204 |
+
def create_video_with_heatmap(video_path, df, mse_embeddings, mse_posture, mse_voice, output_path):
|
205 |
+
from matplotlib.backends.backend_agg import FigureCanvasAgg
|
206 |
+
# Open the video
|
207 |
+
cap = cv2.VideoCapture(video_path)
|
208 |
+
fps = cap.get(cv2.CAP_PROP_FPS)
|
209 |
+
width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
|
210 |
+
height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
|
211 |
+
|
212 |
+
# Create the output video writer
|
213 |
+
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
|
214 |
+
out = cv2.VideoWriter(output_path, fourcc, fps, (width, height + 200)) # Additional 200 pixels for heatmap
|
215 |
+
|
216 |
+
# Prepare the heatmap data
|
217 |
+
heatmap_data = np.vstack((mse_embeddings, mse_posture, mse_voice))
|
218 |
+
|
219 |
+
# Create a figure for the heatmap
|
220 |
+
fig, ax = plt.subplots(figsize=(width/100, 2))
|
221 |
+
im = ax.imshow(heatmap_data, aspect='auto', cmap='YlOrRd')
|
222 |
+
ax.set_yticks([])
|
223 |
+
ax.set_xticks([])
|
224 |
+
plt.tight_layout()
|
225 |
+
|
226 |
+
frame_count = 0
|
227 |
+
total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
|
228 |
+
|
229 |
+
while True:
|
230 |
+
ret, frame = cap.read()
|
231 |
+
if not ret:
|
232 |
+
break
|
233 |
+
|
234 |
+
# Update the heatmap with the current frame position
|
235 |
+
ax.axvline(x=frame_count, color='r', linewidth=2)
|
236 |
+
|
237 |
+
# Convert the matplotlib figure to an image
|
238 |
+
canvas = FigureCanvasAgg(fig)
|
239 |
+
canvas.draw()
|
240 |
+
heatmap_img = np.frombuffer(canvas.tostring_rgb(), dtype='uint8')
|
241 |
+
heatmap_img = heatmap_img.reshape(canvas.get_width_height()[::-1] + (3,))
|
242 |
+
heatmap_img = cv2.resize(heatmap_img, (width, 200))
|
243 |
+
|
244 |
+
# Combine the video frame and the heatmap
|
245 |
+
combined_frame = np.vstack((frame, heatmap_img))
|
246 |
+
|
247 |
+
# Add timecode to the frame
|
248 |
+
timecode = df['Timecode'][frame_count] if frame_count < len(df) else "End"
|
249 |
+
cv2.putText(combined_frame, f"Time: {timecode}", (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 255, 255), 2)
|
250 |
+
|
251 |
+
out.write(combined_frame)
|
252 |
+
frame_count += 1
|
253 |
+
|
254 |
+
# Remove the vertical line for the next iteration
|
255 |
+
ax.lines.pop()
|
256 |
+
|
257 |
+
cap.release()
|
258 |
+
out.release()
|
259 |
+
plt.close(fig)
|
260 |
+
|
261 |
+
return output_path
|