優化 OpenCV 中的邊界框繪圖

Question

def boxDrawing(layerOutput, frameWidth, frameHeight,class_ids,confidences,boxes,img):
    for output in layerOutput:
        for detection in output:
            score = detection[5:]
            class_id = np.argmax(score)
            confidence = score[class_id]
            if confidence > 0.5:
                center_x = int(detection[0] * frameWidth)
                center_y = int(detection[1] * frameHeight)
                width = int(detection[2] * frameWidth)
                height = int(detection[3] * frameHeight)
                left = int(center_x - width / 2)
                top = int(center_y - height / 2)
                class_ids.append(class_id)
                confidences.append(float(confidence))
            boxes.append([left, top, width, height])
    indexes = cv2.dnn.NMSBoxes(boxes, confidences, 0.8, 0.7)
    font = cv2.FONT_HERSHEY_PLAIN
    colors = np.random.uniform(0, 255, size = (len(boxes),3))
    for i in range(len(boxes)):
        if i in indexes:
            x,y,w,h = boxes[i]
            label = str(classes[class_ids[i]])
            confi = str(round(confidences[i],2)) 
            color = colors[i]
            cv2.rectangle(img, (x,y), (x+w,y+h), color,1)
            cv2.putText(img, label+" "+ confi, (x,y+20), font, 1, (255,255,255),1)

這是我將邊界框繪制到檢測的代碼。 但是，當我分析我的代碼時，似乎 np.argmax(score) 消耗了大量時間。 這是用於 boxDrawing function 的 cProfile output：

3375   60.327    0.018  132.575    0.039 2381106525.py:39(boxDrawing)
1    0.424    0.424  742.042  742.042 2381106525.py:68(algorythmYolo)
17010000   10.916    0.000   72.061    0.000 <__array_function__ internals>:177(argmax)
17010000   20.109    0.000   47.640    0.000 fromnumeric.py:1127(argmax)
17010000    7.243    0.000   27.531    0.000 fromnumeric.py:51(_wrapfunc)
17010000   11.303    0.000   58.942    0.000 {built-in method numpy.core._multiarray_umath.implement_array_function}
17010000   17.412    0.000   17.412    0.000 {method 'argmax' of 'numpy.ndarray' objects}

algorythmYolo 是我的主要 function 進行圖像處理； 因此，消耗了大部分時間，並且僅在 boxDrawing function 中調用了 argmax。 有什么方法可以優化 boxDrawing function （因此是 argmax），使其工作得更快？ 提前致謝！

Answer 1

我將假設您的課程處於最后 5 個步驟，這就是為什么您收集最后 5 個索引以獲得 class id 的原因。

一種想法是使用完整的矩陣而不是單獨的一維向量，因為 numpy 已優化為使用矩陣運算：

# Get the matrix of all the output classes
classes = np.array(output)[:,5:]
# class_ids returns a 1d vector with corresponding class
class_ids = np.argmax(classes, axis=1)

這會將您的代碼更改為以下內容：

for output in layerOutput:

    classes = np.array(output)[:,5:]
    class_ids = np.argmax(classes, axis=1)

    for index, detection in enumerate(output):
        confidence = score[class_ids[index]]
        if confidence > 0.5:
            center_x = int(detection[0] * frameWidth)
            center_y = int(detection[1] * frameHeight)
            width = int(detection[2] * frameWidth)
            height = int(detection[3] * frameHeight)
            left = int(center_x - width / 2)
            top = int(center_y - height / 2)
            class_ids.append(class_id)
            confidences.append(float(confidence))
        boxes.append([left, top, width, height])

切片和 maxarg 的示例：

import numpy as np

output = np.array([
    [0,0,0,0,0,0,0,0,2,1,2,3,4],
    [0,0,0,0,0,0,0,0,2,1,2,3,4],
    [0,0,0,0,0,0,0,0,2,1,2,7,4],
    [0,0,0,0,0,0,0,0,2,1,2,3,4],
    [0,0,0,0,0,0,0,0,2,1,6,3,4],
])

print(output)

print(output[:,5:])

classes = output[:,5:]
#[[0 0 0 2 1 2 3 4]
# [0 0 0 2 1 2 3 4]
# [0 0 0 2 1 2 7 4]
# [0 0 0 2 1 2 3 4]
# [0 0 0 2 1 6 3 4]]

print(np.argmax(classes, axis=1))
# [7 7 6 7 5]