Haochen-Wang409 / U2PL

[CVPR'22 & IJCV'24] Semi-Supervised Semantic Segmentation Using Unreliable Pseudo-Labels & Using Unreliable Pseudo-Labels for Label-Efficient Semantic Segmentation
Apache License 2.0
436 stars 61 forks source link

关于两种数据集训练的差异问题 #166

Open syqisme opened 11 months ago

syqisme commented 11 months ago

您好,最近我在您的项目上进行修改并学习,目前在pascal voc上效果较好,但是在cityscapes数据集上效果很差,目前找了很久没找到原因,于是想问您一下,请问您在训练两种数据集上有什么设置上或者训练策略上需要注意的地方呢?目前我看到了一些config文件内的细小差别,可是怎么尝试效果都一般,如果能回复并给我一些建议的话,我将非常感谢!

Haochen-Wang409 commented 11 months ago

请您按照这个 config 进行 Cityscapes 的实验:https://github.com/Haochen-Wang409/U2PL/blob/main/experiments/cityscapes/744/ours/config.yaml

syqisme commented 11 months ago

请您按照这个 config 进行 Cityscapes 的实验:https://github.com/Haochen-Wang409/U2PL/blob/main/experiments/cityscapes/744/ours/config.yaml

十分感谢您的回复,我按照这个去进行尝试,得到结果再来告诉您

lukasingle commented 7 months ago

What is the reason for a single card training error in the windows system question

Haochen-Wang409 commented 7 months ago

This code does not support training on a single GPU.

79zzz commented 2 months ago

local_rank = int(os.environ["LOCAL_RANK"])将local_rank设置成0试试