MoveNetPoseEstimation呈現不準確的關鍵點

我正在嘗試在視頻上運行 MoveNet 姿勢估計模型，但由于某種原因，我的關鍵點非常不準確。我認為這與預測本身沒有任何關系，而是與我如何計算點和繪制然后使用我的估計有關。但是我找不到這些不準確之處來自哪里。

import tensorflow as tf
import numpy as np
from matplotlib import pyplot as plt
import cv2

interpreter = tf.lite.Interpreter(model_path='lite-model_movenet_singlepose_lightning_3.tflite')
interpreter.allocate_tensors()



def draw_keypoints(frame, keypoints, confidence_threshold):
    y, x, c = frame.shape

    shaped = np.squeeze(np.multiply(keypoints, [y,x,1]))
    
    for kp in shaped:
        ky, kx, kp_conf = kp
        if kp_conf > confidence_threshold:
            cv2.circle(frame, (int(kx), int(ky)), 4, (0,255,0), -1) 


cap = cv2.VideoCapture("pushup-stock-compressed.mp4")
while cap.isOpened():
    ret, frame = cap.read()
    
    # Reshape image
    img = frame.copy()
    img = tf.image.resize_with_pad(np.expand_dims(img, axis=0), 192,192)
    input_image = tf.cast(img, dtype=tf.float32)
    
    # Setup input and output 
    input_details = interpreter.get_input_details()
    output_details = interpreter.get_output_details()
    
    # Make predictions 
    interpreter.set_tensor(input_details[0]['index'], np.array(input_image))
    interpreter.invoke()
    keypoints_with_scores = interpreter.get_tensor(output_details[0]['index'])
    
    # Rendering 
    draw_keypoints(frame, keypoints_with_scores, 0.4)
    
    cv2.imshow('MoveNet Lightning', frame)
    
    if cv2.waitKey(10) & 0xFF==ord('q'):
        break
        
cap.release()
cv2.destroyAllWindows()

uj5u.com熱心網友回復：

正如

該示例包含來自網路的一些錯誤預測（看右腿）。

現在對原始影像上的這些關鍵點應用逆仿射變換：

正如我們所看到的，關鍵點繪制在與調整大小 填充影像相同的位置。

完整示例：

import tensorflow as tf
import numpy as np
import cv2

interpreter = tf.lite.Interpreter(
    model_path="lite-model_movenet_singlepose_lightning_3.tflite"
)
interpreter.allocate_tensors()


def draw_keypoints(frame, keypoints, confidence_threshold):
    for kp in keypoints:
        ky, kx, kp_conf = kp
        cv2.circle(frame, (int(kx), int(ky)), 4, (0, 255, 0), -1)


def get_affine_transform_to_fixed_sizes_with_padding(size, new_sizes):
    width, height = new_sizes
    scale = min(height / float(size[1]), width / float(size[0]))
    M = np.float32([[scale, 0, 0], [0, scale, 0]])
    M[0][2] = (width - scale * size[0]) / 2
    M[1][2] = (height - scale * size[1]) / 2
    return M


frame = cv2.imread("gym.png")

# Reshape image
img = frame.copy()
img = tf.image.resize_with_pad(np.expand_dims(img, axis=0), 192, 192)
input_image = tf.cast(img, dtype=tf.float32)

# Setup input and output
input_details = interpreter.get_input_details()
output_details = interpreter.get_output_details()

# Make predictions
interpreter.set_tensor(input_details[0]["index"], np.array(input_image))
interpreter.invoke()
keypoints_with_scores = interpreter.get_tensor(output_details[0]["index"])[0, 0]

img_resized = np.array(input_image).astype(np.uint8)[0]
keypoints_for_resized = keypoints_with_scores.copy()
keypoints_for_resized[:, 0] *= img_resized.shape[1]
keypoints_for_resized[:, 1] *= img_resized.shape[0]
draw_keypoints(img_resized, keypoints_for_resized, 0.4)
cv2.imwrite("image_with_keypoints_resized.png", img_resized)

orig_w, orig_h = frame.shape[:2]
M = get_affine_transform_to_fixed_sizes_with_padding((orig_w, orig_h), (192, 192))
# M has shape 2x3 but we need square matrix when finding an inverse
M = np.vstack((M, [0, 0, 1]))
M_inv = np.linalg.inv(M)[:2]
xy_keypoints = keypoints_with_scores[:, :2] * 192
xy_keypoints = cv2.transform(np.array([xy_keypoints]), M_inv)[0]
keypoints_with_scores = np.hstack((xy_keypoints, keypoints_with_scores[:, 2:]))

# Rendering
draw_keypoints(frame, keypoints_with_scores, 0.4)
cv2.imwrite("image_with_keypoints_original.png", frame)

標籤：Python 张量流 opencv 张量流精简版

上一篇：AttributeError:'builtin_function_or_method'物件沒有屬性'apply'

下一篇：如何打破子行程內的while回圈