search

TencentARC / BrushNet

[ECCV 2024] The official implementation of paper "BrushNet: A Plug-and-Play Image Inpainting Model with Decomposed Dual-Branch Diffusion"
https://tencentarc.github.io/BrushNet/
Other
1.36k stars 114 forks source link

Why not shuffling training dataset #52

Open CharlesGong12 opened 2 months ago

CharlesGong12 commented 2 months ago

Hi, such an excellect work! I am reading your code and find you don't use shuffle when training here. I remember that not using shuffle will damage the convergence effect, from my lessons. Could you explain it? Great Thanks!

xduzhangjiayu commented 2 weeks ago

Hi, such an excellect work! I am reading your code and find you don't use shuffle when training here. I remember that not using shuffle will damage the convergence effect, from my lessons. Could you explain it? Great Thanks!

Hi, Did you solve that when using shuffle=True ? I always got ValueError: DataLoader with IterableDataset: expected unspecified shuffle option, but got shuffle=True Thanks!

CharlesGong12 commented 2 weeks ago

Hi, such an excellect work! I am reading your code and find you don't use shuffle when training here. I remember that not using shuffle will damage the convergence effect, from my lessons. Could you explain it? Great Thanks!

Hi, Did you solve that when using shuffle=True ? I always got ValueError: DataLoader with IterableDataset: expected unspecified shuffle option, but got shuffle=True Thanks!

Nope yet. So I directly use the author's code now.