Code release

[zlaskar]

Dear Authors,

When is the intended date for code release?

Thanks.

[tien-d]

Dear zlaskar,

We plan to release our code no later than November 2022.

Best, Tien

[zlaskar]

Dear Authors,

Thanks for the update. I know it takes time to release code, but is it possible to share the subset of landmark selected for the experiments. For example the 300 landmarks used for training per dataset/scene and hloc-300, hloc-1000, hloc-3000.

Thanks

[tien-d]

@zlaskar,

We released the selected 300 landmarks for the experiments under the landmarks/ folder in each scene. Here is the link to more selected landmarks (3000). Note that since all landmarks are selected based on a greedy fashion, the smaller index landmark has higher score, i.e., the first 300 landmarks (in the selected 3000 landmarks files) are used to train NBE+SLD, the first 1000 landmarks are used for hloc-1000, etc.

[zlaskar]

Dear Authors,

Thanks for the data. That is incredibly helpful. I have hopefully not too many requests remaining :).

The provided link seems to contain the subset of landmarks for 6 scenes. As 7 scenes is a standard dataset to test, is it kindly possible to also get the subset of landmarks for 7 scenes. I am assuming the selected 3D landmarks would correspond to the refined pgt by Eric et. al. that gives the best results in your paper.

Another question if I may, as mentioned in the paper, Section 4 (last paragraph) 'HLoc-LX. In this HLoc variant, we only store X scene landmarks and their descriptors' - as we know one 3D landmark can correspond to multiple 2D local descriptors. Are the number of landmarks equal to the number of local desciptors? If yes, do you just average them?

Thanks in advance.

[lck666666]

Dear zlaskar,

We plan to release our code no later than November 2022.

Best, Tien

hi, your work is really impressive, but when you decide to release the code?

[tien-d]

Hi, the code is now released. Sorry for the delay.