zhiyuan

it is sad because the competition i join is required to submit an offline kernel ,so i cannot connect the internet to downgrad my ipykernel

no,i use the pytorch1.0 with cuda10.0,they are matached

so you mean your code should run in the pytorch0.3?

i find the solution for my problem in stackoverflow, i'm told the reason may be the mismatch between cuda and pytorch,while my cuda and pytorch seem match well for other...

would you mind giving me more deatails about the envs of your code ?

> Sure, you can look at the requirements.txt file to find out the exact dependencies of this project. thank you for reply ,you mean your envs is exactly the source...

i met the same problem ,it seems the code has not use the output of the train_classification,it is puzzled

i have solved the problem

ok

It is clear to understand the match between the concept and the unit using gan dissection, but i have doubts about the match between the location area of the concept...