AICITY2023_track5 多目标跟踪YOLOV8+strongSORT，报错：RuntimeError: CUDA error: device-side assert triggered

YOLOv8已至，精度大涨！教你如何在自定义数据集上训练它 (baidu.com)

(121条消息) 【目标检测】如何使用Yolov8_旅途中的宽~的博客-CSDN博客

ultralytics/ultralytics: NEW - YOLOv8 🚀 in PyTorch > ONNX > CoreML > TFLite (github.com)

ngocnd2402/AICITY2023_HelmetRules (github.com)

注意：标签是从0开始，因为官方给的标签是从1开始的，所以要提前处理一下标签数据

不然会报错：RuntimeError: CUDA error: device-side assert triggered

这个bug主要是标签不匹配的原因，报错的地方未必是真正有问题的地方。

处理标签代码如下：

​

labelfile_path = '/data/aicity_dataset/aicity2023_track5/gt.txt'
new_labeldile_path = '/data/aicity_dataset/aicity2023_track5/gt_new.txt'
m = []
with open(labelfile_path) as f:
    lines = f.readlines()
    for line in lines:
        line = str(line.split(',')[0])+','+ str(line.split(',')[1]) + ','+str(line.split(',')[2])+','+str(line.split(',')[3])+','+ \
        str(line.split(',')[4])+','+str(line.split(',')[5]) + ',' +str(line.split(',')[6])+','+ str(int(line.split(',')[7])-1)+'\n'
        m.append(line)

print(str(m))

with open(new_labeldile_path,'w') as f2:
    f2.writelines(m)

​

 yolo task=detect mode=predict model=/data/CVPR_aicity/runs/detect/train36/weights/best.pt conf=0.7 source=/data/yolov8/test/images save=True