ehsanik / dogtorch Goto Github PK

Hi,
I can't seem to figure out what you did for nearest neighbor for both acting and planning. For acting, the paper says it "uses the features obtained from a Resnet-18 network trained on imagenet." How is that used to predict 5 frames into the future?

Likewise, for planning, nearest neighbor "concatenates the features obtained from resnet-18 trained on imagenet and searches for a plan of the required size that corresponds to the closest feature." If the plan is the sequence of actions in between the first and fifth image, how do you use the features from those images to get that?

I don't think this part was implemented in the code on github.

Thanks!

Hi, I encounter a syntaxerror when I try to run the code

File "/home/zhangdh/dogTorch/training/solver.py", line 49
input = Variable(input.cuda(async=True))
^
SyntaxError: invalid syntax

I am using PyTorch 0.4.0 with anaconda, do you have any ideas about the problem?
I am sorry that the question may be stupid because I am new to PyTorch.
Thanks a lot for the impressive work and the patience!