I would like to ask what is the goal of the network trained using Main.py?

zoubohao / DenoisingDiffusionProbabilityModel-ddpm-

This may be the simplest implement of DDPM. You can directly run Main.py to train the UNet on CIFAR-10 dataset and see the amazing process of denoising.

MIT License

1.48k stars 156 forks source link

I would like to ask what is the goal of the network trained using Main.py? #11

Closed AlanLouis12 closed 1 year ago

AlanLouis12 commented 1 year ago

In the Diffusion.py file I see that loss = F.mse_loss(self.model(x_t, t), x_0, reduction='none') Does this mean that we want to train the Unet network to make the output images closer and closer to the noise? I don't really understand it here.

zoubohao commented 1 year ago

Yes, the model is trained to predict the noise. I recommend you read the paper of DDPM. It is better for you to understand the process of DDPM.