DewarpNet Loader for joint training

We use separate loader for each network. I combined the code, but do i need to fix the loader to bring materials for each network?

Nov 27 '20 01:11 yonghole

yeah, input is the image, load the wc ground-truth, and bm ground-truth.

Nov 27 '20 13:11 sagniklp

What about the alb? I thought alb is also used for recon_loss in texture mapping network. Also, just make total loss 0.5 * wc_loss + 0.5 * bm_loss and back propagate. Right?

Nov 28 '20 08:11 yonghole

Can you provide the code for joint training, thank you very much ！

Feb 18 '21 03:02 Wanwan1996

Sorry for delaying this, will try to do this by the weekend. You can also do it pretty easily, just combine two training codes and rename the variables accordingly. In case you do it, you're welcome to open a pull request, I'll verify & merge.

Mar 04 '21 02:03 sagniklp

I have merged, ok, today I will submit it

Mar 04 '21 03:03 Wanwan1996

Can the code of joint training be released？

Apr 13 '21 08:04 ARnnn

What about the alb? I thought alb is also used for recon_loss in texture mapping network. Also, just make total loss 0.5 * wc_loss + 0.5 * bm_loss and back propagate. Right?

hi, could you please tell me if this way works: just make total loss 0.5 * wc_loss + 0.5 * bm_loss and back propagate so many thx~

Sep 15 '22 09:09 irisXcoding