torch.tensor(images).cuda(async=True)
SyntaxError: invalid syntax
解决方法：
# 将图片和标签转化为tensor
image_var = torch.tensor(images).cuda(non_blocking=True)
label = torch.tensor(target).cuda(non_blocking=True)
模型下载比较慢，可以提前下载：
http://data.lip6.fr/cadene/pretrainedmodels/inceptionv4-8e4777a0.pth
放到C:\Users\用户名\\.cache\torch\hub\checkpoints 下
报错： UserWarning: This DataLoader will create 12 worker processes in total. Our suggested max number of worker in current system is 8 (
cpusetis not taken into account), which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary.
意思是线程设置太多，需要减少
设置如下
# 进程数量，最好不要超过电脑最大进程数。windows下报错可以改为workers=0
workers = 8
可是又会报错
AttributeError: Can’t pickle local object ‘main..TrainDataset’
所以我们修改为：
workers = 0
CUDA out of memory. Tried to allocate 72.00 MiB (GPU 0; 4.00 GiB total capacity; 2.75 GiB already allocated; 9.74 MiB free; 2.77 GiB reserved in total by PyTorch)
先修改：
# 设定GPU ID
os.environ["CUDA_VISIBLE_DEVICES"] = '0'
因为我们确实只有一块GPU，并且显存确实爆了
所以我们需要修改：
# 小数据集上，batch size不易过大。如出现out of memory，应调小batch size
batch_size = 6 # 原来是24，这个根据你的显卡调，我的是1050ti
可以看到几乎拉满了
训练开始，然后报错：
line 2111, in _verify_batch_size raise ValueError(“Expected more than 1 value per channel when training, got input size {}”.format(size))
ValueError: Expected more than 1 value per channel when training, got input size torch.Size([1, 1536])
查看记录得知，代码这一句报错，原因是train_loader最后一张无法训练
train(train_loader, model, criterion, optimizer, epoch)
所以我们设置train_loader，加上参数：
drop_last=True
train_loader = DataLoader(train_data, batch_size=batch_size, shuffle=True, pin_memory=True, num_workers=workers, drop_last=True)
