carlsummer
carlsummer
have you solved the problem? I met the same trouble
maybe conda install pytorch torchvision cudatoolkit=10.2 -c pytorch
def is_parallel(model): return type(model) in (nn.parallel.DataParallel, nn.parallel.DistributedDataParallel) msd = model.module.state_dict() if is_parallel(model) else model.state_dict() # model state_dict
python -m torch.distributed.launch --nproc_per_node=2 tools/train.py
它是多卡默认，但是你是单卡在跑，所以报错了 It is multi-card default, but you are running with a single card, so an error is reported
I tried 512 * 512 and 64 * 512, because the graph for my requirements is 400 * 3000, so it is better to test 64 * 512. The line...
I didn't compare the effects of different sizes on wireframe datasets. But you can try, training should not be too long, I because the company has the need to detect...
 
I saw several issues in the issue about how to train my own data set. The task my company now gives me is also to train my own data set,...
@sachinkaundal https://github.com/carlsummer/python_developer_tools That is a development tool I wrote. There are many tool functions in it. You can copy it,