PBA Distance Transform using NPP returns non-exact EDT.

[jeanchristopheruel]

Related to this topic I followed the examples at https://github.com/NVIDIA/CUDALibrarySamples/tree/master/NPP/distanceTransform In these examples, I set the first row a sites.

// nppiDistanceTransformPBA_8u32f_C1R_Ctx
[0]: 0 0 0 0 0 0 0 0 0 0
[1]: 1 1 1 1 1 1 1 1 1 1
[2]: 1 1 1 1 1 1 1 1 1 1
[3]: 3 3 3 3 3 3 3 3 3 3
[4]: 3 3 3 3 3 3 3 3 3 3
[5]: 4 4 4 4 4 4 4 4 4 4
[6]: 5 5 5 5 5 5 5 5 5 5
[7]: 6 6 6 6 6 6 6 6 6 6
[8]: 7 7 7 7 7 7 7 7 7 7
[9]: 8 8 8 8 8 8 8 8 8 8
[10]: 9 9 9 9 9 9 9 9 9 9

// nppiDistanceTransformAbsPBA_8u16u_C1R_Ctx
[0]: 0 0 0 0 0 0 0 0 0 0
[1]: 1 1 1 1 1 1 1 1 1 1
[2]: 1 1 1 1 1 1 1 1 1 1
[3]: 3 3 3 3 3 3 3 3 3 3
[4]: 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5
[5]: 4.5 4.5 4.5 4.5 4.5 4.5 4.5 4.5 4.5 4.5
[6]: 5.5 5.5 5.5 5.5 5.5 5.5 5.5 5.5 5.5 5.5
[7]: 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5
[8]: 7.5 7.5 7.5 7.5 7.5 7.5 7.5 7.5 7.5 7.5
[9]: 8.5 8.5 8.5 8.5 8.5 8.5 8.5 8.5 8.5 8.5
[10]: 9.5 9.5 9.5 9.5 9.5 9.5 9.5 9.5 9.5 9.5

[mkhadatare]

Hi @jeanchristopheruel,

Thank you for reporting this issue. Could you please provide a small reproducible example along with details of your testing environment, including information about the GPUs, CUDA Toolkit version, and any other relevant configurations?

[soooch]

Hey @mkhadatare I'm seeing the exact same issue on one system, but not on another. I am using nppiDistanceTransformPBA_8u32f_C1R_Ctx.

System with issue: RTX 4070 CUDA/NPP version: 11.8

System without issue: Jetson Orin Nano 4GB CUDA/NPP version: 11.4

The nature of the issue is exactly as demonstrated in @jeanchristopheruel's post. Locations which should have distances of 2 are reported to have distances of 1. This appears to right itself at distance 3. Then at positions which should have distance 4, we see distance 3.5. Then the rate of the distance increments correctly, though the values themselves are incorrect. (3.5, 4.5, 5.5 where it should be 4, 5, 6)

[soooch]

This is internally tracked at 4832970 btw.

[jeanchristopheruel]

Woohoo thanks a lot for the followup!

[mkhadatare]

Hi @soooch Thank you for reporting this. We are currently working on the issue under bug tracker ID 4832970 and will keep you updated on the progress.