about train

[a13483685]

about train

[usamahjundia]

i would suggest loading the state dict and print the keys of the weights. Most probably since the authors used multigpu training (dataparallel module) , the pretrained weights keys have something like "module." prepended to the keys. Probably the less hacky way to do this is to use the same dataparallel module.

I think for your case, you can just load the state dict, iterate over the keys and values, and write another state dict with the corrected keys. THen you load the corrected state dict. This is a bit hacky but it worked for me.