pytorch1.8复现TSN踩坑记录（如何在源代码中进行过更改）

1 运行main.py时，

1.1 报错RuntimeError: Legacy autograd function with non-static forward method is deprecated.

参考1，针对TSN网络
参考2
参考3

新版pytorch中的前向传播forward变成静态的了
（1）basic_opt.py中最下面的forward中加入.forward(input)

(2)models.py中的base——model处加入.forward

1.2 报错RuntimeError: view size is not compatible with input tensor’s size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(…) instead.

修改main.py下的，将view替换成reshape

1.3 报错IndexError: invalid index of a 0-dim tensor. Use tensor.item() in Python or tensor.item() in C++ to convert a 0-dim tensor to a number

将main.py中的loss.data[0]改为loss.data，同理更改prec1和prec5

1.4 报错RuntimeError: invalid argument 5: k not in range for dimension at /pytorch/aten/src/THC/generic/THCTensorTopK.cu:24

将main.py中topk的5修改为3

2 运行test_models.py时

2.1 net = torch.nn.DataParallel(net.cuda(devices[0]), device_ids=devices)报错AssertionError: Invalid device id

指定训练显卡

2.2RuntimeError: CUDA out of memory. Tried to allocate 192.00 MiB (GPU 0; 9.78 GiB total capacity; 7.50 GiB already allocated; 190.38 MiB free; 7.72 GiB reserved in total by PyTorch)

根据报错位置

加入with torch.no_grad():