search

Visual-Attention-Network / SegNeXt

Official Pytorch implementations for "SegNeXt: Rethinking Convolutional Attention Design for Semantic Segmentation" (NeurIPS 2022)
Apache License 2.0
794 stars 85 forks source link

Random seeds variance #48

Open baize97 opened 1 year ago

baize97 commented 1 year ago

When I try the SegNeXt training with different random seeds, I find that the mIoU will have a large variance (up to 1% to 2%).

Tiny image

Base image

How do you think such impact on the comparison with other models and the reproducibility of results?

MenghaoGuo commented 1 year ago

Yes, it depends on benchmark. In all dataset, ADE20K dataset has the largest variance. If you want to get stable result, trying to use cityscapes.