Hi, is your pretrained model the same as in original paper? both trained on Visual Gnome?

[youngfly11]

Hi, is your pretrained model the same as in original paper? both trained on Visual Gnome?

[violetteshev]

I just transferred the pretrained model provided in original repo (structure and weights) to PyTorch.

[youngfly11]

ok, thank you very much , i just want to confirm that whether the results produced by your code are the same as the original papers?

[violetteshev]

I tested the code on a random subset of MSCOCO images and checked that the extracted features are the same as features extracted by original code, but I didn't check the detection performance on any dataset (although it should be the same).

[youngfly11]

thanks

[jbdel]

I just transferred the pretrained model provided in original repo (structure and weights) to PyTorch.

Hello violette,

Could you please tell me how you managed to do that ?

[violetteshev]

I manually copied all the weights from one model to another.