Time and size of dataset to train the model detecting planar surface with Mask R-CNN

[sonnguyen9800]

Hello NVlabs,

I am trying to reproduce your project using custom dataset. However, my goal is slightly simplified as my only need is to detect planar surface from RGB pictures of cubes, cylinder and so on ... , by Mask R-CNN.

Given that you guys mention the effectiveness of the neuron network to detect planar surface. I want to ask about the number of images you used to train the model. As I see, the number of class is 2 (planar and non-planar). How much time it takes to properly distinguish the two?

Thanks in advance and wish you all best healthy these days.

Son

[ajithvcoder]

@sonnguyen9800 its not related to u r issue . Were u able to build nms and roialign ? which cuda version u r using ?

[sonnguyen9800]

@sonnguyen9800 its not related to u r issue . Were u able to build nms and roialign ? which cuda version u r using ?

Sorry to cause any misunderstanding, as English is not my first language. I have not built the nms and roialign yet. But, i want to prepare custom dataset first. So, i want to ask about the minimum number of dataset needed for the Mask R-CNN.

Best Regads,