shaofengzeng/SuperPoint-Pytorch

color input

Closed this issue · 3 comments

Hello, thanks for your promising job. would you has planed to implemented rgb color instead of gray input.

I think it is easy to change the input gray image to a color one. The main problem is to modify the image augmentation methods in photometric_augmentation.py, you can refer to this repo. ,or write these methods with the help of imgaug. Some other methods, like ratio_preserving_resize , are also need minor modifications.

thanks for you reply. i has rewrite this repo to support rgb input. the training process convergence quick and has a more lower loss value. but the evaluation result is quite bad.

thanks for you reply. i has rewrite this repo to support rgb input. the training process convergence quick and has a more lower loss value. but the evaluation result is quite bad.

I suggest checking the input data. Make sure they are consistent with the training stage.