the number of keypoints for matching

[KN-Zhang]

The work is inspiring.

Note that we uniformly sample a dense grid of keypoints on both the depth map and the image.

So how many points are sampled for matching?

[HpWang-whu]

Hi @KN-Zhang , Thanks for your interest! we sample 32*44 = 1408 points for indoor data and 32*80 = 2560 points for outdoor data.

Yours,

[KN-Zhang]

Hi! What's the difference between .color.gtd.png and .depth.gtd.png in ScanNet? Maybe the latter is the depth of the downsampled point cloud? :))

And, may I ask the difference between .depth.gtd.png and .depth.png?

[HpWang-whu]

Hi @KN-Zhang, .depth.png is the depth map projected from the sparse (down-sampled) point cloud and used for registration. .color.gtd.png and .depth.gtd.png are the ground truth dense depth maps (projected from a dense point cloud) corresponding to the .color.png and *.depth.png for correspondence evaluation.

Yours,

[KN-Zhang]

Hi @KN-Zhang, .depth.png is the depth map projected from the sparse (down-sampled) point cloud and used for registration. .color.gtd.png and .depth.gtd.png are the ground truth dense depth maps (projected from a dense point cloud) corresponding to the .color.png and *.depth.png for correspondence evaluation.

Yours,

Thanks for your patient response. 👍