yolov7 _pickle.UnpicklingError: STACK_GLOBAL requires str 问题修复
问题现象
在启动训练时候报错了
1 Model Summary: 415 layers, 37201950 parameters, 37201950 gradients, 105.1 GFLOPS 2 3 Scaled weight_decay = 0.0005 4 Optimizer groups: 95 .bias, 95 conv.weight, 98 other 5 Traceback (most recent call last): 6 File "train.py", line 610, in <module> 7 train(hyp, opt, device, tb_writer) 8 File "train.py", line 248, in train 9 image_weights=opt.image_weights, quad=opt.quad, prefix=colorstr('train: ')) 10 File "/home/yolo系列/yolov7/utils/datasets.py", line 78, in create_dataloader 11 prefix=prefix) 12 File "/home//yolo系列/yolov7/utils/datasets.py", line 392, in __init__ 13 cache, exists = torch.load(cache_path), True # load 14 File "/home/qfz/anaconda3/envs/yolov7/lib/python3.7/site-packages/torch/serialization.py", line 608, in load 15 return _legacy_load(opened_file, map_location, pickle_module, **pickle_load_args) 16 File "/home/qfz/anaconda3/envs/yolov7/lib/python3.7/site-packages/torch/serialization.py", line 777, in _legacy_load 17 magic_number = pickle_module.load(f, **pickle_load_args) 18 _pickle.UnpicklingError: STACK_GLOBAL requires str
问题分析：
回想起是偷懒直接用yolov5训练过的数据集拷贝到yolov7下进行重新训练结果报错。
最终解决方法：
经过波斯大佬的指点，把训练数据集下面的 labels.cache文件删除即可。
