LuChengTHU / dpm-solver

Official code for "DPM-Solver: A Fast ODE Solver for Diffusion Probabilistic Model Sampling in Around 10 Steps" (Neurips 2022 Oral)
MIT License
1.57k stars 120 forks source link

problem with order 2 CIFAR-10 (VP deep continuous-time model ) #53

Open molisi opened 7 months ago

molisi commented 7 months ago

hello,thanks to your wonderful work!i run your code with CIFAR-10 (VP deep continuous-time model ),dpms-solver order2 recently, i try so many times but still get wrong fid with the order2(table 6) result,could u give me some help such as the implement of hyperparameter? thank u so much!!!! please save the child ~~~

LuChengTHU commented 7 months ago

Hi @molisi , yes table 6 is based on the first version of DPM-Solver. When I proposed DPM-Solver++, I found a better way to interpolate between time steps so I discarded the implementation for the first version, which causes a little mismatch with the original dpm-solver paper; though the FID numbers are still reasonable.

If you want to exactly reproduce the FID numbers in my original paper, please check this version of example: https://github.com/LuChengTHU/dpm-solver/tree/0aa673cb00cb8c6beddd37074364b7a3c4dbbfff/example_v1