Linfeng-Tang / SeAFusion

The code of " Image fusion in the loop of high-level vision tasks: A semantic-aware real-time infrared and visible image fusion network"
MIT License
184 stars 34 forks source link

关于后续分割任务的标签问题. #19

Open JasonWong30 opened 1 year ago

JasonWong30 commented 1 year ago

你好,作者。针对于您这份论文里面提到的分割任务的ground truth,我提出一个疑问。在这论文中您提到这个工作是联合训练的,然后又在论文中提到“in the image fusion field it is difficult to provide the ground truth of fused images for training a high-level vision task model“,那么您这个ground truth是基于哪个图像做出来的?如何得到的?期待您的答复!谢谢啦。

Linfeng-Tang commented 1 year ago

正如论文中提到的我们使用的这个数据集是基于MFNet数据集构造的,其中使用的的GT也是该数据集提供的。据我的了解 这个数据集的GT应该是基于可见光图像得到的,可能其中也考虑了一些红外图像的信息 但是并没有完全考虑。

JasonWong30 commented 1 year ago

非常感谢您的回复。我这里还有有个疑问就是您在消融实验中所提到的单阶段联合训练是什么,与你论文中使用的联合训练有何不同?

Linfeng-Tang commented 1 year ago

就是 直接同时更新融合和分割网络的参数

JasonWong30 commented 1 year ago

好的,十分感谢您的回复!!!