Artefacts with applying effects on certain letters about msdfgen HOT 7 CLOSED

Keep in mind that this is an approximate rendering method, and it is designed to be most precise at glyph edges. The further away from an edge, the higher the probability that the distance is not evaluated precisely, resulting in an artifact such as this one. This may be mitigated by increasing the distance field density.

from msdfgen.

I am experiencing similar issues with msdf-atlas-gen version 1.9. I believe this is a regression, as I can render fonts created with prior versions without these artefacts. The screenshot below shows the issue with a screenPxRange of 1.8000001907348633. Note the n and u. These artefacts appear and dissappear depending as the image grows and shrinks (e.g. not all screenPxRange values).

You can replicate my images with:

msdf-atlas-gen.exe -font Roboto-Regular.ttf -charset chars.txt -pxrange 2 -dimensions 256 256 -format png -json fnt.json -imageout fnt.png

Where chars.txt is a text file with the following characters

"\"\\ ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz1234567890!`?'.,;:()[]{}<>|/@^$-%+=#_&~*"

from msdfgen.

The issue can also be observed by looking at the bitmaps created by the new (defective) and prior (good) versions (generated with same commands as above, but with -dimensions 512 256. Note that two colors touch at full brightness (255) in the new version. Therefore, the interpolated median value precisely between these pixels can be high.

Prior versions:

Latest version:

from msdfgen.

@neurolabusc You are right that this is a new regression and it would seem that I've made an embarassing mistake in the new error correction algorithm, which should now be fixed by 35caf38. Please try it out and let me know if there are still problems, if not I will release a hotfix version.

from msdfgen.

@Chlumsky thanks for the quick response. I will not have access to a development system for the next week. Is it possible that you could share a compiled copy - macOS, Linux or Windows would be fine. You could even use the "pre-release" function for Github releases to let users test your fix.

from msdfgen.

@neurolabusc Alright, I have published the pre-release, so you can test it.

from msdfgen.

1. The pre-release resolves my regression issue.
2. I suggest you close this issue, as the original issue can be resolved by generating the recent RGBA MTSDF textures.
3. I only tested the 32-bit application, as Windows 10 rejected the 64-bit pre-release claiming it exhibits the Wacatac.B!ml script virus. I suspect this is a false alarm, but it might be nice to see if the 64-bit version can be packaged in such a way to avoid this issue.

from msdfgen.