Thanks for your great job! I have one quick question here.
I noticed that when training the votenet, the code sets shuffle=True for TEST_DATALOADER(which is very weird to me). That means every time for inference, we used the same val list in a different order.
I have thought it should be fine because we set net.eval(). However, the results actually performs very differently under the different order of val list. May I ask why this happened?
And also, if I get a very high result under one specific order of val list(but drop a lot under the original order of val list), can I report it as the final result?
Hi authors,
Thanks for your great job! I have one quick question here.
I noticed that when training the votenet, the code sets
shuffle=True
for TEST_DATALOADER(which is very weird to me). That means every time for inference, we used the same val list in a different order.I have thought it should be fine because we set
net.eval()
. However, the results actually performs very differently under the different order of val list. May I ask why this happened?And also, if I get a very high result under one specific order of val list(but drop a lot under the original order of val list), can I report it as the final result?