#6 Tiling errors evident in PNG output with large solid areas

closed
nobody
None
5
2010-01-29
2010-01-24
Bill Kalarness
No

Hi, Your fix for ID: 2934646 did improve the quality of the renders (5400 x 5400), but didn't totally solve the problem. From my previous bug, it sounds like you are saying there are limitations and there need to be realistic expectations for the output quality. I agree 100%! :) However, these problems still persist and are visible in my images (scaled to fit the screen, or seen at actual size.) Granted, the actual size output does diminish the effect of the seams somewhat - but they are still very visible, without having to hunt for them.

I'm not sure how you can quantify a reasonable viewing distance for on-screen display. (?) I do understand that you are trying to set expectations. :)

Please have a look at the screenshot I've included. This is a cut of one of my images - scaled to fit the screen. There are big solid areas in the render, where the tile seams make it appear that the image was segmented incorrectly. Not saying that this is what's happened, it's just how I'm describing the output I'm viewing. It's hard giving good examples, given the attachment limitations of sourceforge. My previous example was not great, but I didn't have time to come up with something better.

Thanks for your time & effort... much appreciated!!

-Bill

Discussion

  • Bill Kalarness
    Bill Kalarness
    2010-01-24

    screen capture of tiling error (cut then, scaled to fit screen)

     
    Attachments
  • Bill Kalarness
    Bill Kalarness
    2010-01-24

    both images in this flam3 have large solid areas

     
    Attachments
  • Bill Kalarness
    Bill Kalarness
    2010-01-24

    This is using v.76

     
  • I have uploaded a test version with a different overlapping tiling approach. It does show some tiling image registration misalignment (a few pixels off) but on two of my monitors does not show the visible seams that are so visible on your test images. There is a step in the rendering process called Density Estimation that was causing the visible seams. This approach creates tiles that are 8 bytes wider on each side, so the Density estimation (which uses a 7 byte radius) is not affected. On tiling reassembly, the extra overlap is discarded.

    I am seeing misalignment on very sharp lines but not in other areas. It looks better than before but is still not perfect.

    It is a test executable, so its not the installer package.

    Please let me know how this works with your other images.

    https://sourceforge.net/projects/flam4/files/flam4Mac/TilingTest/Flam4CUDA.app.zip/download

     
    • status: open --> closed
     
  • Further enhancements have been made to tiling in Version 0.77. Tiles now overlap slightly and the overlaps between two tiles are blended in the final image. On the example flame file posted for this bug, the tile seams are now very difficult to find.

    Fixed in version 0.77.