Trained models used in the paper

[fdila]

Hi, i noticed that the pretrained model you made available outputs a vector with 32 features, while the registration tests shown in the supplementary material used feature vectors with higher dimensionality. Did you notice any differences in performances using a higher dimensionality? In that case, could you make those pretrained models available? Thank you

[yewzijian]

Hi, the registration results are all based on 32D features. We did experiment with larger number of dimensions but that yielded diminishing improvements (see Fig. 3 in main paper for feature matching performance). The supplementary material are also based on the same 32D feature descriptors.

Unfortunately, I no longer have the checkpoints for the other dimensionalities.

[fdila]

Thanks