Changing zeros to Nans for better missing detection marking

Author: John Harrington

demo/demo.py

@@ -281,15 +281,15 @@ def main():
                             draw_pose(kpt,image_bgr) # draw the poses
                     else:
                         if keypoints is None:
-                            keypoints = np.array([[[0, 0]]*len(COCO_KEYPOINT_INDEXES)])
+                            keypoints = np.array([[[np.nan, np.nan]]*len(COCO_KEYPOINT_INDEXES)])
                         else:
-                            keypoints = np.append(keypoints, [[[0, 0]]*len(COCO_KEYPOINT_INDEXES)], axis=0)
+                            keypoints = np.append(keypoints, [[[np.nan, np.nan]]*len(COCO_KEYPOINT_INDEXES)], axis=0)
             else:
                 #Fill undetected frames with zero vectors
                 if keypoints is None:
-                    keypoints = np.array([[[0, 0]]*len(COCO_KEYPOINT_INDEXES)])
+                    keypoints = np.array([[[np.nan, np.nan]]*len(COCO_KEYPOINT_INDEXES)])
                 else:
-                    keypoints = np.append(keypoints, [[[0, 0]]*len(COCO_KEYPOINT_INDEXES)], axis=0)
+                    keypoints = np.append(keypoints, [[[np.nan, np.nan]]*len(COCO_KEYPOINT_INDEXES)], axis=0)
 
             if args.showFps:
                 fps = 1/(time.time()-last_time)