Skip to content

无法复现论文中的精度 #103

Description

@Asuna88

作者您好,我看到有人提出相似的问题,但是没有找到答案。

我采用batchsize 8 iter60000 初始0.01学习率的策略,但是最终只能在cityscapes val上到大约67%,68%的miou(STDC1-50)。
我使用的训练脚本就是github README.MD中的训练脚本。

请问我是不是哪里使用错误了么?为什么原封不动的使用您的代码,却无法达到(或者接近)您论文中的mIOU指标?至少相差6-7个百分点)

希望您能给点建议,非常感谢!

训练脚本如下:

export CUDA_VISIBLE_DEVICES=0
python -m torch.distributed.launch \
--nproc_per_node=1 train.py \
--respath checkpoints/train_STDC1-Seg/ \
--backbone STDCNet813 \
--mode train \
--n_workers_train 12 \
--n_workers_val 1 \
--max_iter 60000 \
--use_boundary_8 True \
--pretrain_path checkpoints/STDCNet813M_73.91.tar

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions