Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.


Messages - Cyrille Damez

Pages: 1 ... 113 114 [115] 116
1711
Yes, that would be fine. Thank you.

1712
It looks like a difference in the way we compute/read the tangents and binormals. This is going to change in the next version of Designer, and it will likely fix that problem. To be completely sure, would it be possible for you to share these meshes (or other simple ones which have the same problem), so that I can check that the problem is indeed fixed ?

1713
Hey, good job !
Regarding the "curious result" with paraboloid and HDR files, I suspect the code we use to handle this file format in the player applies some sort of gamma correction, which is almost like taking the square root of the luminosity (typically taking it to the power 1/2.2, whereas the square root is 1/2). Since the formula for the paraboloid is 1-x²-y² and the formula for the half-sphere is just the square root of that, that may explain what you are observing.

1714
I am not familiar with maxscript, but I think you can use file names with spaces in your case if you "protect" the variables containing the file names with double-quotes :
Code: [Select]
" specialization --input " + templatePath +
                        " --presets-path " + resourcesPath +
                        " --output-name \"" + sbs +
                        "\" --output-graph-name " + RemoveSpaces(sbs) +
                        " --output-path " + sbsPath;

I haven't checked, but maybe templatePath, ressourcePath and sbsPath need protecting also. Obviously, RemoveSpaces(sbs) should not. ;)

HTH

1715
It's feasible to do a 3x3 median filter, although the result is probably going to be slow compared to Photoshop's filter. The basic idea is to use an FxMap that draws square patterns the size of a pixel. The color of each pattern should be decided by a function graph which reads all colors in the 3x3 neighbourhood and computes the median. You can read pixel values from an FxMap input in a function graph using the Sample node.
In theory, you could do a median filter of any fixed size (5x5, 7x7, etc.) but anything larger than 3x3 is going to be tedious to plug because the size of the comparison network necessary to compute the median grows with the amount of pixels to compare.

1716
Yes it looks like a bug. Could you tell us which version of Designer you installed, and what OSX version you are running ? Was it the Steam version ?

1717
Can you try using the D3D10 engine instead of the default D3D9 engine and tell us if it fixes the issue ? You can choose which engine Substance Designer uses in the Tools->Switch Engine... menu.

1718
Thanks for the help.
Which tutorial was that ?

1719
The timings given for each node are measured on the machine you are running the Substance Editor on, probably using the DX9 engine if you are on a PC and have not changed the default settings. It is quite likely that the actual performance on a given mobile target will be slower as your computer's GPU is probably much more powerful than any existing mobile's CPU.
However, the relative speed of each node should be similar, so you can still use those timings as an indication of where optimisations are needed.
It is possible that you will get performances closer (i.e. slower) to what you will observe on your mobile target in Designer if you switch to the CPU (SSE2) engine (in Tools->Switch Engine...).

1720
Quote
Are there any specific nodes I shouldn't be using?
The graph editor should indicate which nodes take the most rendering time in red. In general, noises(FxMaps) are the slowest, warps and blurs come second (the directional warps and directional blurs should be faster).
Although the actual timing depends on the final hardware you will be rendering on (any mobile cpu is likely to render slower than your pc's gpu or cpu) the relative "weight" of nodes should be similar.
Quote
I have found that using "Color Input" causes big performance hits.
Strange. That should not be the costliest part of your Substance. Passing a color input texture to the Substance engine only makes a copy of the texture pixels. How did you measure that, and how large is that hit ?
Quote
Also if I have a SVG connected to a blend node with an opacity of 0 (set in code) do you still redraw that SVG?
In order not to draw the SVG, your blend node should use the "switch" blending mode instead of "copy" (unless you are working in greyscale, in which case "copy" and "switch" behave the same, and the svg will not be drawn).


1721
Ok, we may have found what's going on (but we are still unable to reproduce it). It seems to be a weird floating point rounding difference.

Could you test the attached substance and tell us whether it suffers from the same bug or not ? (hopefully not)

1722
Also, could you try to switch to a different engine (in Tools->Switch Engine...), and tell us if you see any improvement. Thanks.

1723
Thanks. We had one report of that problem, but it was on a mac running an old version of OSX, and we never managed to reproduce the issue.
What's your CPU model/brand ?

Also, could you try to switch to the SSE2 engine (in Tools->Switch Engine...) and tell us if you still get the same result ?

1724
Can you tell us more about your configuration (Substance Designer version, OS, graphics card, graphics driver, etc.) ?
Thanks.

1725
This error shows up whenever a security program is denying Substance Designer the right to identify the machine it's running on. For example, this happens if you have Comodo Internet Security installed. You just need to tell Comodo (or whatever security enforcing program you are running) that Substance Designer is a trusted program so that it can access the COM interface.

Pages: 1 ... 113 114 [115] 116