Pretrained weights

[muaz-urwa]

Thank you for the awesome project. It would be awesome if you can also share the pretrained weights (checkpoints) with it.
Would it be possible ?

[andrewjong]

I need to retrain them because I was simply prototyping SwapNet before, and deleted them by mistake. But note the implementation is only 90% finished. There are still two things to implement/fix, namely (1) Copy the face data in the final result (issue #6), and (2) debug the perceptual loss to achieve high texture quality (issue #7). A released model at this time would not have those included. Would you like a model with this status or do you prefer to wait for a final product?

I'm also currently busy until mid-December (working on grad school applications). The earliest I can promise is late December.

[muaz-urwa]

Dear Andrew,
I would really appreciate it if you can share what you have right now.

Good luck for the grad school application !

[Larkinnjm1]

Hey folks, thanks once again for uploading this unbelievable repo. Did you get any joy uploading your training weights?

[andrewjong]

Hi all, thank you for your patience! And thank you @Larkinnjm1 for the reminder. I just started training a model for upload. Going to let it run overnight, then upload tomorrow evening.

Proof:

[andrewjong]

Update: models have finished training. Download here and extract under the project root: https://drive.google.com/open?id=1kaeHlGf9h3vZLBxr4D5yQUgToxEUWuMh

Please note that these models are far from perfect. I have yet to properly implement the Perceptual Loss, which is crucial for proper texture transfer. See #7 .

Let me know if there are any issues!