Rémi Pautrat

Hi, maybe there are simply no keypoint detected. Did you visualize the images and the predictions of the network?

Did you modify anything from the code online?

Hi, these changes should not be a problem in theory. So I don't know why you don't get keypoints on the second image. This is indeed a very weird behavior......

Hi, if I remember correctly, it is because the checkpoints also contain the state_dict of the optimizer for different checkpoints. Thus, if the training of the model trained on VIDIT...

Hi, in `superpoint/configs/magic-point_shapes.yaml`, you can add the following entry: ``` data: generation: split_sizes: training: 5000 ``` But I don't understand why 10000 would be too much for your GPU... The...

Hi, Unfortunately, there was not easier choice than just experimenting until I could find the best parameters. I cannot guarantee that the current values are optimal, but they should work...

I don't see any reason for changing $\lambda$ and $\lambda_d$ when the training set is harder. The only reasons to change them would be the following I think: - $\lambda$:...

Hi, unfortunately not. I am not maintaining this repo anymore, so it stayed at the stage of Tensorflow 1. In case you really need to use TF2 and decide to...

Hi, We did slightly modify the original descriptor loss, as this formulation gave us better results. I cannot help you much with your code, as it is a reimplementation and...

I am closing this for now, feel free to reopen if you need additional assistance.