Training SuperGlue on custom data

[Shuhul24]

According to the comment by @sarlinpe (which suggested to use LightGlue), I need help in how to train SuperGlue on my own custom dataset instead of MegaDepth data that is used in this repo. @rpautrat @Phil26AT

[RobinHan24]

how to finetune my own dataset and how make our own data, thanks a lot.

[guipotje]

It depends on what your dataset has available as labels for supervision. If correspondence can be obtained (from depth & poses, or as optical flow), i think it is straightfoward, just need to adapt the data loaders. If no correspondence is available, the only option I see is to use the augmentation part only with raw images, which is also straightforward to adapt in the data loaders.

[fighterzzzh]

how to finetune my own dataset and how make our own data, thanks.

[Zhao-Shi-jie]

how to finetune my own dataset and how make our own data, thanks.

Hi! I'd like to know if you've already fixed this issue，thanks

[Zhao-Shi-jie]

According to the comment by @sarlinpe (which suggested to use LightGlue), I need help in how to train SuperGlue on my own custom dataset instead of MegaDepth data that is used in this repo. @rpautrat @Phil26AT

Hi! I'd like to know if you've already fixed this issue，thanks

[swjtuer0762]

根据（建议使用 LightGlue）的评论，我需要帮助如何在我自己的自定义数据集上训练 SuperGlue，而不是在此存储库中使用的 MegaDepth 数据。

你好！我想知道你是否已经解决了这个问题，谢谢

您好！我也有相同的问题，能够留下联系方式互相交流吗？我的邮箱是[email protected]. I also have the same problem, can I leave my contact information to communicate with each other? My email is [email protected].