ByteTrack传入检测结果报错：TypeError元组索引类型错误

问题 #

无法将YOLOv5的检测结果传入ByteTrack，按官方指引尝试后失败，代码如下：

# Perform object detection with YOLOv5
results = model(frame)
#print(results)

# Process detection results
bbs = []
for det in results.pandas().xyxy[0].iterrows():
    _, detection = det
    xmin = float(detection.xmin) 
    ymin = float(detection.ymin) 
    xmax = float(detection.xmax - xmin) 
    ymax = float(detection.ymax - ymin) 
    confidence = float(detection.confidence)
    class_id = int(detection.name)
    
    if class_id >= 0 and class_id < len(class_labels):
            class_label = class_labels[class_id]
    if confidence > conf_threshold: # 0.4
        # Append detection to the list of bounding boxes
        bbs.append((xmin, ymin, xmax, ymax, confidence, class_id)) 
        #print(bbs)
        
dets = (bbs, num_classes)
#print(dets)

# Pass detection results to BYTETracker
online_targets = tracker.update(dets, img_size) #, info_imgs

报错信息：

TypeError                                 Traceback (most recent call last)
Cell In[17], line 77
     73 dets = (bbs, num_classes)
     74 #print(dets)
     75 
     76 # Pass detection results to BYTETracker
---> 77 online_targets = tracker.update(dets, img_size) #, info_imgs
     79 # Process tracking results
     80 for target in online_targets:

File ~\AppData\Local\anaconda3\envs\bytetrackpypi\lib\site-packages\bytetracker\byte_tracker.py:186, in BYTETracker.update(self, dets, _)
    183 lost_stracks = []
    184 removed_stracks = []
---> 186 xyxys = dets[:, 0:4]
    187 xywh = xyxy2xywh(xyxys)
    188 confs = dets[:, 4]

TypeError: tuple indices must be integers or slices, not tuple

解决方案 #

问题根源 #

1. 传入ByteTrack的dets格式错误：ByteTrack的update方法需要Numpy数组格式的检测结果，你却传了一个包含列表和num_classes的元组，导致代码尝试对元组做数组切片操作时报错。
2. 边界框格式错误：你把xmax和ymax计算成了宽高（xmax - xmin、ymax - ymin），但ByteTrack需要的是原始的xmax、ymax坐标，即(xmin, ymin, xmax, ymax)格式。

修改步骤 #

1. 先导入Numpy（如果未导入）：

import numpy as np

2. 修正边界框计算逻辑：
   将代码中这两行：

xmax = float(detection.xmax - xmin) 
ymax = float(detection.ymax - ymin) 

修改为：

xmax = float(detection.xmax) 
ymax = float(detection.ymax) 

3. 转换检测结果为Numpy数组，无需打包num_classes：
   将：

dets = (bbs, num_classes)

修改为：

# 若有检测结果则转成Numpy数组，无结果则传空数组
dets = np.array(bbs) if bbs else np.empty((0, 6))

4. 正常调用ByteTrack的update方法：

online_targets = tracker.update(dets, img_size)

说明 #

ByteTrack的update方法要求dets是形状为(N,6)的Numpy数组，每一行对应一个检测框的xmin, ymin, xmax, ymax, confidence, class_id，修改后即可符合要求，解决报错。

内容的提问来源于stack exchange，提问作者Yiannis