About depth map of SGD dataset

[ader47]

Hi! @wenquanlu I‘ve noticed that the accuracy of the depth map in Static Gesture Dataset (SGD) is meter level. During training, should we use the ground truth of the depth map or the depth map converted from the normal map? Thanks! 🙏

[wenquanlu]

Hi! we used the ground truth of depth map provided in the SGD.

There is one preprocessing step though, We first segment out the hand depth map and apply normalization to the depth map. On a scale of 0 (black) to 1 (white), the closest surface is given the value of 1.0, and the furthest surface is given the value of 0.2.

Hope this helps.

[ader47]

Hi! we used the ground truth of depth map provided in the SGD.

There is one preprocessing step though, We first segment out the hand depth map and apply normalization to the depth map. On a scale of 0 (black) to 1 (white), the closest surface is given the value of 1.0, and the furthest surface is given the value of 0.2.

Hope this helps.

OK, thank you so much!

[wenquanlu]

This is the procedure to process training set for RHD. SGD is very similar to this