Wilbur Precipitation-Based Erosion adding weird bumps

[Rubikia]

Hi there,

I've been using Wilbur quite extensively and had a lot of fun with the programme and it truly is powerful and able to do a lot. I don't know the ins and outs of it, but I know how at least to do what I want with it thanks to Waldronate's help

Recently I have been using it to erode a large heightmap I created with Inkscape with 9 layers, and it worked well for it when I did it at A1 scale. I have increased the size of the heightmap layers in hopes of having a 20,000 x 14,127 pixel sized map, but this is where my Wilbur renders have been running into problems.

I set the offset of the first layer of mountains this time to 100, and the mountains between 100 and 400, then setting the fill value of the rivers to -1, and the height clip low=0.01 and high=10000000. I then generated noise at 100% using the river mask, and then applied 25 passes using the land mask. After that, I selected the rivers again and inverted, setting its fill to -1. Then I applied 10% noise using the river mask, set the height clip again, and ran 25 passes using the land mask, and repeated these settings afterwards but with 5% noise.

This worked well aside from about 1 quarter of the map having weird bumps across it. Unfortunately, I only noticed this at the very end of several day-long process of rendering.

Wilbur Relief Render.PNG

I tried again, completing the erosion cycle but the same bumps are appearing.

Wilbur Capture.PNG

And right now, the bumps are starting to appear once more during the operation (this time I set it to 30 passes in the vain hope that would change something).

Wilbur Capture2.PNG

I suspect that these bumps must be what is left of the percentage noise, but I have no idea why they are not being eroded or why its a very particular area of the map.

I would greatly appreciate any help or advice with this issue.

Cheers,

[Naima]

Hi there,

I've been using Wilbur quite extensively and had a lot of fun with the programme and it truly is powerful and able to do a lot. I don't know the ins and outs of it, but I know how at least to do what I want with it thanks to Waldronate's help

Recently I have been using it to erode a large heightmap I created with Inkscape with 9 layers, and it worked well for it when I did it at A1 scale. I have increased the size of the heightmap layers in hopes of having a 20,000 x 14,127 pixel sized map, but this is where my Wilbur renders have been running into problems.

I set the offset of the first layer of mountains this time to 100, and the mountains between 100 and 400, then setting the fill value of the rivers to -1, and the height clip low=0.01 and high=10000000. I then generated noise at 100% using the river mask, and then applied 25 passes using the land mask. After that, I selected the rivers again and inverted, setting its fill to -1. Then I applied 10% noise using the river mask, set the height clip again, and ran 25 passes using the land mask, and repeated these settings afterwards but with 5% noise.

This worked well aside from about 1 quarter of the map having weird bumps across it. Unfortunately, I only noticed this at the very end of several day-long process of rendering.

Wilbur Relief Render.PNG

I tried again, completing the erosion cycle but the same bumps are appearing.

Wilbur Capture.PNG

And right now, the bumps are starting to appear once more during the operation (this time I set it to 30 passes in the vain hope that would change something).

Wilbur Capture2.PNG

I suspect that these bumps must be what is left of the percentage noise, but I have no idea why they are not being eroded or why its a very particular area of the map.

I would greatly appreciate any help or advice with this issue.

Cheers,

3x3 blurr pass.

[Rubikia]

3x3 blurr pass.

Thanks for the advice, I tried applying the 3x3 blur after applying noise (and beforehand too in another attempt), but unfortunately the strange bumps/less eroded area is showing on the right again.

Wilbur Relief Render 2.PNG

I'm going to let it render a bit more, but it is strange that Wilbur is eroding certain areas differently.

[Greason Wolfe]

Are you doing a basin fill before the erosion process? If not, it may take a while (100s of passes on precip perhaps) before that bit of roughness eventually disappears. That doesn't necessarily mean that there isn't another problem of some sort contributing to this issue. Perhaps Waldronate will be by in the near future and will have better advice, suggestions, and insight into the issue.

[waldronate]

Those could be caused by any number of problems, including getting close to a cycle in the random number generator. I added a median morphological filter in the last release to deal with that kind of noise, but the erode morphological filter will also work. I don't recommend using blur because it kills all of the fine detail that you waited so long for.
I don't have anywhere near the amount of patience required to do a full-resolution surface like that, so I tend to do multiresolution processing. If I ever get around to doing any significant work on Wilbur again, I'd probably add an image pyramid mode to simplify a lot of processing. However, I expect that large-data terrain models are going to completely obsolete the kind of image processing things that Wilbur does in the next few years.

[Rubikia]

Using the morphological filter worked very well in getting rid of those bumps! Thank you! I set it to 2 using a 8bit png of the last complete render of it, but unfortunately at the moment it is largely killing/flattening some of the detail and making the contours far more visible. I'm not too sure why it is doing this, but I suppose it is because it's generally flattening the image?

Wilbur Relief Render 3.PNG

Wilbur Relief Render 4.PNG

[Naima]

Using the morphological filter worked very well in getting rid of those bumps! Thank you! I set it to 2 using a 8bit png of the last complete render of it, but unfortunately at the moment it is largely killing/flattening some of the detail and making the contours far more visible. I'm not too sure why it is doing this, but I suppose it is because it's generally flattening the image?

Wilbur Relief Render 3.PNG

Wilbur Relief Render 4.PNG

it looks like stepped , is this result of that filter?

[waldronate]

Which morphological filter did you use? Dilate, Median, and Erode correspond to a local maximum, most likely, and local minimum, respectively (blur is arithmetic mean, if that helps). For Median, the Window Size value is the radius of the square window around the current pixel to perform the operation (1=3x3 window, 2=5x5 window, 3=7x7 window, and so on). The system grabs all pixels in the local neighborhood, sorts the pixels by value, and then picks the value from the middle of the sorted array. Dilate and Erode get the maximum or minimum value in the local 3x3 window once for each value of Samples specified, which is very broadly equivalent to working on a Samples radius square window around the current pixel.

In an ideal world, Wilbur's Dilate, Median, and Erode morphological operations would be implemented using the same internals (collect array, process array, output resulting value to temporary surface), but that's not how the evolution happened. It would be nice if those operations (and the analogous windowed mean/blur operation) would work with window shapes other than square (a round kernel, for example, would give edges a different character than the current square one). Having the NaN killer for all of the morphological operations would be nice, but its presence in Median was purely a result of laziness on my part because I never got around to doing a unique NaN killer and the thing I was working on kept pulling NaNs.

Normally, a Window Size of 1 with the median filter should be sufficient to get rid of single-pixel noise because it will get the most likely value from the 3x3 neighborhood centered on each pixel. Using a larger window will reshape the terrain something vaguely like a blur. Median will get rid of the extreme high and low values in its window area, so it will indeed have the effect of flattening the image, but with a slightly different character than blur because median will change the overall amount of energy in the window, while blur will keep the same energy but spread it over the local window area.

Adding some more specialized filters like the Kuwahara filter (a kind of adaptive noise reduction filter that doesn't kill the edges as much) might be a nice exercise some year, but it's even harder to explain the motivation for that kind of filter in the context of terrain processing.

[waldronate]

it looks like stepped , is this result of that filter?

The steps should be a result of the original data being an 8-bit PNG. Median won't change those flats or edges very much, but it will kill very fine detail like single-pixel noise items. The larger the window on the median filter, the larger the size of individual details that it's likely to kill. I added it to kill those single-pixel towers that tend to crop up in my preferred blur/basin fill/incise/precipiton loop. Using Morphological Erode also kills the towers, but it widens river valleys and sharpens mountain peaks as well; Dilate will broaden the towers into little square blocks, and Median will just sort of smooth it over without killing detail in the same way that blur would.

[johnvanvliet]

when i use wilbur i use 32 bit floating point *.bt images that have pixel values between 0 and 65536
this seams to give me the best results

"gdal_translate" converts the tiff images into bt format for use in wilbur