how to visualize the reconstruction result

[YixuannnL]

Maybe a bit silly question is how you visualize your reconstruction result when your autoencoder training is done?
In my understanding maybe i should take the input to the decoder, which is the original image added noise, and using a noise scheduler denoise the input once. Just like
latents = self.noise_scheduler.step(noise_pred, timesteps, dec_x).prev_sample
gen_img = self.image_decoder(latents)
Is it reasonable?

Or should i denoise the noisy input of decoder iteratively in a for loop, using the noise predicted by the decoder? But i only add noise to the original image once, so i dont think this is reasonable maybe