Pretrained baseline weights

[Hayoung93]

@NaJaeMin92
Hi, I read FixBi interestingly and want to reproduce some results.

Can you provide .pt files?

FixBi/src/utils.py

Lines 36 to 42 in fdf370d

	def load_net(args, net, head, classifier):
	print("Load pre-trained baseline model !")
	save_folder = args.baseline_path
	net.module.load_state_dict(torch.load(save_folder + '/net.pt'), strict=False)
	head.module.load_state_dict(torch.load(save_folder + '/head.pt'), strict=False)
	classifier.module.load_state_dict(torch.load(save_folder + '/classifier.pt'), strict=False)
	return net, head, classifier

(net.pt, head.pt, classifier.pt)

You mentioned this repo for pretrained weights, but to me, it seems two repo has different network architecture (so cannot be loaded even if I train using the mentioned repo).

Should I edit networks' architecture of DANN (as same as this repo), train it on the Office-31 dataset, and use it as a pretrained baseline weight?

Thank you for your work and I will wait for your response.

[Arsiuuu]

Hi, have you found the pretrained models?

[Hayoung93]

@Eureka-JTX Unfortunately no. I'd excluded loading part and tried to train from scratch but failed (I assume pretrained weights are nessasary, note that authors wrote we start to train our networks with pretrained baseline weights at the paper).

I could try training those weight for my own, however I have less motivation to do such work, so currently trying to reproduce other domain adaptation papers' results.

[Arsiuuu]

@Hayoung93 Thanks for your patient answer! I've reproduced the MSTN from DSBN(https://github.com/wgchang/DSBN), but I only obtained the weights of encoder and discriminator rather than head and classifier in this repo, so the result is even lower than DSBN.

[xixi1998]

no pretrained model weights. so sad.

[toshi2k2]

The results of this work seem non - reproducible.