search

KiseKloset / DM-VTON

👗 DM-VTON: Distilled Mobile Real-time Virtual Try-On
https://sites.google.com/view/ltnghia/research/DMVTON
Other
107 stars 23 forks source link

Reproduce result in paper #11

Closed npk7264 closed 10 months ago

npk7264 commented 10 months ago

I have trained by using a file notebook, but in stage 2 of Teacher, it's running out of GPU memory. I reduced batch_size from 14 to 12. However, FID ~ 60. Could you let me know your config, GPU, and parameters when training to have FID ~ 28 as paper? Thank you.

zero-nnkn commented 10 months ago

In Colab, we demo how to train the model and the used resource. So, we set the total epochs for each stage as 2 (niter=1 and niter_decay=1). If you want to reproduce, please follow the report in the paper using 100 epochs (niter=50 and niter_decay=50) for each stage.