如何修复YOLOv3处理MP4时出现的IndexError: invalid index to scalar variable？

问题描述 #

使用YOLOv3处理MP4视频时，持续触发以下错误：

invalid index to scalar variable

错误指向获取YOLO输出层名称的代码段，原代码如下：

import numpy as np
import cv2

# initialize minimum probability to eliminate weak predictions
p_min = 0.5

# threshold when applying non-maxia suppression
thres = 0.

# 'VideoCapture' object and reading video from a file
video = cv2.VideoCapture('parking1.mp4')

# Preparing variable for writer
# that we will use to write processed frames
writer = None

# Preparing variables for spatial dimensions of the frames
h, w = None, None

# Create labels into list
with open('coco.names') as f:
    labels = [line.strip() for line in f]
# Initialize colours for representing every detected object
colours = np.random.randint(0, 255, size=(len(labels), 3), dtype='uint8')

# Loading trained YOLO v3 Objects Detector
# with the help of 'dnn' library from OpenCV
# Reads a network model stored in Darknet model files.
network = cv2.dnn.readNetFromDarknet('yolov3.cfg',
                                     'yolov3.weights')

# Getting only output layer names that we need from YOLO(ERRORRRRRRRRRR HEREEE) 
ln = network.getLayerNames()
ln = [ln[i[0] - 1] for i in network.getUnconnectedOutLayers()]

# Defining loop for catching frames
while True:
    ret, frame = video.read()
    if not ret:
        break

    # Getting dimensions of the frame for once as everytime dimensions will be same
    if w is None or h is None:
        # Slicing and get height, width of the image
        h, w = frame.shape[:2]

    # frame preprocessing for deep learning
    blob = cv2.dnn.blobFromImage(frame, 1 / 255.0, (416, 416),
                                 swapRB=True, crop=False)

错误截图显示代码执行到ln = [ln[i[0] - 1] for i in network.getUnconnectedOutLayers()]时抛出异常。

错误原因 #

不同OpenCV版本中，getUnconnectedOutLayers()的返回值类型不一致：

• 旧版本返回标量数组（如array([200, 227, 254])）
• 新版本返回二维数组（如array([[200], [227], [254]])）

原代码中i[0]的写法仅适配新版本，若使用旧版OpenCV，会因标量无法索引而报错。

修复方案 #

统一处理两种返回值类型，修改获取输出层名称的代码：

ln = network.getLayerNames()
# 处理不同版本的返回值格式
out_layers = network.getUnconnectedOutLayers()
# 判断返回值是标量还是数组
if isinstance(out_layers[0], list):
    ln = [ln[i[0] - 1] for i in out_layers]
else:
    ln = [ln[i - 1] for i in out_layers]

完整优化代码 #

以下是可直接运行的完整YOLOv3视频处理代码，包含检测、绘制结果、保存输出视频的完整流程：

import numpy as np
import cv2

# 参数配置
p_min = 0.5  # 最小置信度阈值
thres = 0.3  # NMS非极大值抑制阈值
input_video = 'parking1.mp4'
output_video = 'parking_detected.mp4'

# 加载视频
video = cv2.VideoCapture(input_video)
if not video.isOpened():
    print(f"无法打开视频文件: {input_video}")
    exit()

# 初始化视频写入器
writer = None
h, w = None, None

# 加载类别标签
with open('coco.names', 'r') as f:
    labels = [line.strip() for line in f.readlines()]
# 生成类别颜色
colours = np.random.randint(0, 255, size=(len(labels), 3), dtype='uint8')

# 加载YOLOv3模型
network = cv2.dnn.readNetFromDarknet('yolov3.cfg', 'yolov3.weights')
# 设置使用GPU加速（可选，需CUDA版本OpenCV）
# network.setPreferableBackend(cv2.dnn.DNN_BACKEND_CUDA)
# network.setPreferableTarget(cv2.dnn.DNN_TARGET_CUDA)

# 获取输出层名称（修复版本）
ln = network.getLayerNames()
out_layers = network.getUnconnectedOutLayers()
if isinstance(out_layers[0], list):
    ln = [ln[i[0] - 1] for i in out_layers]
else:
    ln = [ln[i - 1] for i in out_layers]

# 逐帧处理视频
while True:
    ret, frame = video.read()
    if not ret:
        break

    # 获取帧尺寸
    if w is None or h is None:
        h, w = frame.shape[:2]

    # 帧预处理
    blob = cv2.dnn.blobFromImage(frame, 1/255.0, (416, 416), swapRB=True, crop=False)
    network.setInput(blob)
    outputs = network.forward(ln)

    # 存储检测结果
    boxes = []
    confidences = []
    class_ids = []

    # 解析输出结果
    for output in outputs:
        for detection in output:
            scores = detection[5:]
            class_id = np.argmax(scores)
            confidence = scores[class_id]
            if confidence > p_min:
                # 转换为原始帧坐标
                box = detection[0:4] * np.array([w, h, w, h])
                centerX, centerY, width, height = box.astype('int')
                x = int(centerX - (width / 2))
                y = int(centerY - (height / 2))
                boxes.append([x, y, int(width), int(height)])
                confidences.append(float(confidence))
                class_ids.append(class_id)

    # 非极大值抑制去除重复框
    indices = cv2.dnn.NMSBoxes(boxes, confidences, p_min, thres)

    # 绘制检测结果
    if len(indices) > 0:
        for i in indices.flatten():
            x, y, w_box, h_box = boxes[i]
            color = colours[class_ids[i]].tolist()
            label = f"{labels[class_ids[i]]}: {confidences[i]:.2f}"
            # 绘制边框
            cv2.rectangle(frame, (x, y), (x + w_box, y + h_box), color, 2)
            # 绘制标签背景
            cv2.rectangle(frame, (x, y - 20), (x + len(label)*10, y), color, -1)
            # 绘制标签文本
            cv2.putText(frame, label, (x, y - 5), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0,0,0), 2)

    # 初始化视频写入器
    if writer is None:
        fourcc = cv2.VideoWriter_fourcc(*'mp4v')
        writer = cv2.VideoWriter(output_video, fourcc, 30, (frame.shape[1], frame.shape[0]), True)

    # 写入处理后的帧
    writer.write(frame)

    # 实时显示（可选，按q退出）
    cv2.imshow('YOLOv3 Video Detection', frame)
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

# 释放资源
video.release()
if writer is not None:
    writer.release()
cv2.destroyAllWindows()

注意事项 #

1. 确保yolov3.cfg、yolov3.weights、coco.names文件路径正确，可从YOLO官方仓库下载
2. 若运行缓慢，可开启GPU加速（需安装CUDA版本的OpenCV）
3. 调整p_min和thres参数可平衡检测精度与速度

内容的提问来源于stack exchange，提问作者Maryam