问题解答
问题一：OSError: [Errno 12] Cannot allocate memory
参考：
死亡Error：OSError: [Errno 12] Cannot allocate memory
OSError: [Errno 12] Cannot allocate memory. But memory usage is actually normal
执行PyTorch程序，发生内存不足错误
- 内存查询
监视内存，查看是否是内存不足
# 打开两个窗口，分别查看CPU内存和显卡内存
# 每隔1秒查询一次
$ watch -n 1 free -m
$ wathc -n 1 nvidia-smi
- num_workers
确实不是因为内存不足，那么修改DataLoader的num_workers为0，再重新运行即可
num_workers (int, optional): how many subprocesses to use for data
loading. ``0`` means that the data will be loaded in the main process.
(default: ``0``)
问题二：Process finished with exit code 137 (interrupted by signal 9: SIGKILL)
前几天突然遇到这个问题：
Process finished with exit code 137 (interrupted by signal 9: SIGKILL)
程序启动后被系统KILLED，在网上找了资料，发现是说内存不足的问题
查询内存使用情况，发现大多数内存都被缓存占据了
$ free -h
total used free shared buff/cache available
Mem: 62G 13G 450M 308M 49G 48G
Swap: 7.6G 4.2G 3.4G
- 额外阅读
Process finished with exit code 137 in PyCharm
multiprocessing.set_start_method(method)
问题三：RuntimeError: CUDA error: initialization error
参考：RuntimeError: CUDA error: initialization error
不可在DataLoader或DataSet内将任何数据放到CUDA上，而是等到程序运行出DataLoader之后（也就是到了train里的时候）将数据放到CUDA上。
问题四：RuntimeError: invalid argument 0: Sizes of tensors must match except in dimension 0
调用DataSet的__getitem__方法返回的image数据维度应该一致