search

zhang-haojie / wesam

[CVPR 2024] Code for "Improving the Generalization of Segmentation Foundation Model under Distribution Shift via Weakly Supervised Adaptation"
MIT License
130 stars 9 forks source link

有关 weak_aug和strong_aug #27

Open wangjing60755 opened 1 month ago

wangjing60755 commented 1 month ago

代码中的weak_aug采用flip，strong_aug采用色彩增强，这个操作有啥理论支撑吗？在各数据集中都是有效的吗？

我的数据对颜色比较敏感，在监督训练中使用色彩增强效果较差。

对于色彩敏感的任务，strong_aug应该怎么设置？是否有可分享的经验？

感谢！

zhang-haojie commented 1 month ago

In semantic perception tasks, there is less experience to refer to. However, for classification, there are many self-training methods available. One notable reference is the "End-to-End Semi-Supervised Object Detection with Soft Teacher" for object detection.