Open kayser-soze opened 4 years ago
Hello , I installed pip install mxnet-cu100==1.5.1, and run demo.py successfully,but when I run train.py,after training. I met the error.
Traceback (most recent call last):
File "experiments/fgfa_rfcn/fgfa_rfcn_end2end_train_test.py", line 20, in
can anyone tell me how to fix it? thanks very much!
Just set allow_extra=True.
Just set allow_extra=True.
Thanks for your help.but when I tried to change allow_extra=allow_extra with allow_extra=True in mxnet\module\base_module.py line 699,the problem is still there. then I also change
def set_params(self, arg_params, aux_params, allow_missing=False, force_init=True,
allow_extra=False):
with
def set_params(self, arg_params, aux_params, allow_missing=False, force_init=True,
allow_extra=True):
in mxnet\module\base_module.py line 671,
the problem is still there .
Just set allow_extra=True.
Thanks for your help.but when I tried to change allow_extra=allow_extra with allow_extra=True in mxnet\module\base_module.py line 699,the problem is still there. then I also change def set_params(self, arg_params, aux_params, allow_missing=False, force_init=True, allow_extra=False): with def set_params(self, arg_params, aux_params, allow_missing=False, force_init=True, allow_extra=True): in mxnet\module\base_module.py line 671, the problem is still there . 您这个问题解决了 我遇到了同样的问题 希望得到您的帮助,感谢。
Just set allow_extra=True.
这个问题有什么好的解决方法吗?谢谢回复
TL;DR: you can install mxnet using pip instead of compile it from source!
For all the ones having trouble to complie mxnet from source:
I did not manage to compile mxnet from source and encountered several problems mentioned in other issues. Then I tried to compile a newer version of mxnet (official 1.5.1). Doing so I noticed that the operator files already exist in the official mxnet source. Thus, I installed this mxnet version using pip - I have Cuda 10.0 so I've used the following command line:
pip install mxnet-cu100==1.5.1
and it works!
After that, I had out of memory problems while running demo.py. I have GTX 1080 with 8GB of RAM. I dealt with it by changing KEY_FRAME_INTERVAL from 9 to 5 (in fgfa_rfcn/demo.py), and changing the code in demo.py as in https://github.com/msracver/Flow-Guided-Feature-Aggregation/issues/16#issuecomment-444388962