The current model configuration doesn't work well on the line dataset because the timestep is not large enough.

tanelp / tiny-diffusion

A minimal PyTorch implementation of probabilistic diffusion models for 2D datasets.

628 stars 52 forks source link

The current model configuration doesn't work well on the line dataset because the timestep is not large enough. #4

Open bopan3 opened 1 year ago

bopan3 commented 1 year ago

As the ablation study in readme.md shows, the current model configuration doesn't work well on the line dataset. The reason is the the timestep is not large enough such that Xt is not in standard gaussian distribution. If we set timestep as 250 instead of 50, the result is nice.

toannguyen1904 commented 8 months ago

As the ablation study in readme.md shows, the current model configuration doesn't work well on the line dataset. The reason is the the timestep is not large enough such that Xt is not in standard gaussian distribution. If we set timestep as 250 instead of 50, the result is nice.

Helpful!