How to fix ‘_pickle.UnpicklingError: invalid load key, ‘＜‘ ‘ error in Pytorch

最新推荐文章于 2024-08-31 16:29:17 发布

Rory·Shaw

最新推荐文章于 2024-08-31 16:29:17 发布

阅读量1.7k

点赞数

文章标签： pytorch 深度学习 python

本文链接：https://blog.csdn.net/weixin_42335814/article/details/129705455

版权

The problem I encountered when I ran the official code of maskrcnn-benchmark for facebookresearch,which was wrong when loading the pre-training model.

The code runs on a remote server at the school and the graphics card is an NVIDIA P100.

checkpointer = DetectronCheckpointer( cfg, model, optimizer, scheduler, output_dir, save_to_disk) extra_checkpoint_data = checkpointer.load(cfg.MODEL.WEIGHT) arguments.update(extra_checkpoint_data)

I expect to run the code correctly and understand why this is the problem.

原因：The reason about the problem is that the previous download was not finished. So when I deleted the original file and re-downloaded it, the problem was solved.
参数没有办法反序列化，长度小于标准，可能的原因是参数没有下载完整，需要重新下载参数。

参考：https://stackoverflow.com/questions/56391392/how-to-fix-pickle-unpicklingerror-invalid-load-key-error-in-pytorch