Drawing landmarks on live webcam feed

[Eric Bubar]

I'm having trouble getting the new mediapipe to draw landmarks on the live webcam feed. I think i've got the new API up and running and detecting live camera keypoints, but really am stuck on getting the points and lines to display. I'm running the following in a Jupyter notebook where i've downloded the lite model as that's the only link I can find for the updated solutions on the documentation. Open to any suggestiong on how to code a function to get the keypoints and lines to display like they did with the draw_landmarks function from mediapipe of old! 

import mediapipe as mp
from mediapipe.tasks import python
from mediapipe.tasks.python import vision
import numpy as np
import cv2

#Setup a model with options for live streaming
model_path = 'pose_landmarker_lite.task'
BaseOptions = mp.tasks.BaseOptions
PoseLandmarker = mp.tasks.vision.PoseLandmarker
PoseLandmarkerOptions = mp.tasks.vision.PoseLandmarkerOptions
PoseLandmarkerResult = mp.tasks.vision.PoseLandmarkerResult
VisionRunningMode = mp.tasks.vision.RunningMode

# Create a pose landmarker instance with the live stream mode:
def print_result(result: PoseLandmarkerResult, output_image: mp.Image, timestamp_ms: int):
    print('pose landmarker result: {}'.format(result))

options = PoseLandmarkerOptions(
    base_options=BaseOptions(model_asset_path=model_path),
    running_mode=VisionRunningMode.LIVE_STREAM,
    result_callback=print_result)

#Open webcam
#Setup computer vision for live webcam
cap = cv2.VideoCapture(0) # uncomment for live webcam video analysis
width = cap.get(cv2.CAP_PROP_FRAME_WIDTH) # Set width of camera
height = cap.get(cv2.CAP_PROP_FRAME_HEIGHT) # Set heigh of camera
font = cv2.FONT_HERSHEY_SIMPLEX #Set font for computer vision

with PoseLandmarker.create_from_options(options) as landmarker:
  while cap.isOpened(): # as long as the webcam is open this will happen
      ret, frame = cap.read() # Saves the image from your webcam as a frame    
    # OPENCV reads in as BGR.  This line recolors image to RGB
      image = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
      image.flags.writeable = False # Saves memory by making image not writeable    
      frame_timestamp_ms = int(cap.get(cv2.CAP_PROP_POS_MSEC))  
    # Recolor image back to BGR for OPENCV
      image.flags.writeable = True
      image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
      numpy_frame_from_opencv = np.array(image)
      mp_image = mp.Image(image_format=mp.ImageFormat.SRGB, data=numpy_frame_from_opencv)
      pose_landmarker_result = landmarker.detect_async(mp_image, frame_timestamp_ms)
      cv2.imshow('test', image)

      if cv2.waitKey(10) & 0xFF == ord('q'): # This puts you out of the loop above if you hit q
        break
   
cap.release() # Releases the webcam from your memory
cv2.destroyAllWindows() # Closes the window for the webcam

[Niko Avradopoulos]

+1 looking for a migrated draw landmarks solution!