Issue of training via 3 stages

[lumos1iu]

Hello Author. I don't understand why we need to use 3 stage training. And stage 2 and stage 3 seem to be the same for the model. I am appreciated for your answer!

[lumos1iu]

Can't encoder and decoder be trained together?

[Afsalts]

were u able to train it @lumos1iu

[lumos1iu]

were u able to train it @lumos1iu

I tried, but the losses connot converge