Degradation of the quality of the displayed model (colors) in pyvista

[3rav]

When I change the time step or anything that forces the replacement of result values ​​(after about 8 changes) there is a degradation in the quality of the displayed model (colors) in pyvista only for remote rendering.

Before:

After:

After a large number of such changes, the model becomes completely unclear.

[jourdain]

What do you mean by only in "remote rendering"? Are you saying that the mesh stay clean when using local rendering?

When you use the regular native rendering of PyVista, do you see the same behavior?

[bourdaisj]

Thanks for the reproduction example and for the confirmation - we'll take a look

[bourdaisj]

@3rav Your screenshots looks like you are using local rendering. But you mention "only for remote rendering" in your first message? It is a confusion/error on your side? Asking this because it seems we are reproducing the problem in local rendering.
For the record:

• local rendering: serialized geometry/topology/scalar/vectors sent to the client -> browser based, VTK.js, WebGL rendering
• remote rendering: VTK OpenGL rendering on the server, with images being streamed to the browser over the websocket

Kindly comment on this

[3rav]

My apologies for the confusion and the error in the title. You are absolutely right, I am using local rendering.

To clarify, the geometry is being rendered on the client's GPU to ensure smooth interaction and fast rotations, while avoiding unnecessary load on the server's CPU/GPU. I mixed up the terminology in my initial message.

Thank you for pointing this out, and I am sorry for the mix-up.

"only for remote rendering" - I meant not on the server

[bourdaisj]

Gotcha, no worries. Thanks for the confirmation - We'll try taking a look in the following days

[3rav]

Any info?

[3rav]

Please find attached a simplified example where the aforementioned problem occurs:
example.py

Results example .vtu and .frd (pyvista/pyvista#8255)
Results_vtu_frd.zip

.vtu

.frd

[3rav]

Any suggestions

[jourdain]

I could not iterate through timesteps with the example and data file. So I could not reproduce the issue.

[3rav]

In this example, you only need to change the color resolution ("Colors" slider) several times:

New example files for time (vtu & pvd):
all.zip

For .frd I remind that you should use the development version of pyvist (dev 0.48)

[3rav]

What about these examples?

[jourdain]

I seems to be working fine on my end with vtk 9.6 on my mac.

What is your runtime? Is it possible that you are having some graphic driver issue?

[3rav]

I tested on 2 machines (both have this problem):

--------------------------------------------------------------------------------
  Date: Thu Apr 09 04:27:56 2026 UTC

                OS : Linux (Ubuntu 25.10)
            CPU(s) : 16
           Machine : x86_64
      Architecture : 64bit
       Environment : Python
        GPU Vendor : AMD
      GPU Renderer : AMD Radeon Graphics (radeonsi, gfx1200, LLVM 20.1.8, DRM 3.64, 6.17.0-20-generic)
       GPU Version : 4.6 (Core Profile) Mesa 25.2.8-0ubuntu0.25.10.1
     Render Window : vtkXOpenGLRenderWindow
  MathText Support : True

  Python 3.13.7 (main, Mar  3 2026, 12:19:54) [GCC 15.2.0]

           pyvista : 0.48.dev0
               vtk : 9.6.0
             numpy : 2.4.3
        matplotlib : 3.10.8
            scooby : 0.11.0
             pooch : 1.9.0
            pillow : 12.1.1
 typing-extensions : 4.15.0
          cyclopts : 4.10.0
             scipy : 1.17.1
              tqdm : 4.67.3
             trame : 3.12.0
      trame-client : 3.11.2
      trame-server : 3.10.0
         trame-vtk : 2.10.3
     trame-vuetify : 3.2.0
     nest-asyncio2 : 1.7.2
--------------------------------------------------------------------------------

and:

--------------------------------------------------------------------------------
  Date: Thu Apr 09 04:38:10 2026 UTC

                OS : Linux (Ubuntu 24.04)
            CPU(s) : 4
           Machine : x86_64
      Architecture : 64bit
       Environment : Python
        GPU Vendor : Intel
      GPU Renderer : Mesa Intel(R) Graphics (ADL-N)
       GPU Version : 4.6 (Core Profile) Mesa 25.2.8-0ubuntu0.24.04.1
     Render Window : vtkXOpenGLRenderWindow
  MathText Support : True

  Python 3.12.3 (main, Mar  3 2026, 12:15:18) [GCC 13.3.0]

           pyvista : 0.48.dev0
               vtk : 9.6.1
             numpy : 2.4.4
        matplotlib : 3.10.8
            scooby : 0.11.0
             pooch : 1.9.0
            pillow : 12.2.0
 typing-extensions : 4.15.0
          cyclopts : 4.10.1
              tqdm : 4.67.3
             trame : 3.12.0
      trame-client : 3.11.4
      trame-server : 3.10.0
         trame-vtk : 2.11.6
     trame-vuetify : 3.2.1
     nest-asyncio2 : 1.7.2
--------------------------------------------------------------------------------

I need to run a test on these machines but on Windows.

[3rav]

Version running on Windows 11,after playing around with what should be displayed, this is what happened (degradation):

or:

the legend can change between changes in its resolution (font color, size, position of the inscription):

after reloading the page (F5):

[3rav]

Ubuntu 22.04 and NVIDIA @jourdain :

--------------------------------------------------------------------------------
  Date: Thu Apr 09 16:57:20 2026 UTC

                OS : Linux (Ubuntu 22.04)
            CPU(s) : 8
           Machine : x86_64
      Architecture : 64bit
       Environment : Python
        GPU Vendor : NVIDIA Corporation
      GPU Renderer : Quadro K1100M/PCIe/SSE2
       GPU Version : 4.5.0 NVIDIA 470.256.02
     Render Window : vtkXOpenGLRenderWindow
  MathText Support : True

  Python 3.10.12 (main, Mar  3 2026, 11:56:32) [GCC 11.4.0]

           pyvista : 0.48.dev0
               vtk : 9.6.1
             numpy : 2.2.6
        matplotlib : 3.10.8
            scooby : 0.11.0
             pooch : 1.9.0
            pillow : 12.1.1
 typing-extensions : 4.15.0
          cyclopts : 4.10.1
              tqdm : 4.67.3
             trame : 3.12.0
      trame-client : 3.11.4
      trame-server : 3.10.0
         trame-vtk : 2.11.6
     trame-vuetify : 3.2.1
--------------------------------------------------------------------------------

[jourdain]

I could not run with the time dependent dataset, your application is not working. So I just played with the number of colors and I could not make it happen. You will need to provide a better example if you want us to look into it.

Also the issue is not related to trame. It seems to be related to rendering with VTK on AMD/Intel.

[3rav]

example_v2.zip

[jourdain]

No need to tag me... I see the thread.

Based on what I can see, the artifact seems to come from a messed up lookup table and/or mesh definition.
This is not related to trame if using remote rendering. And if the issue is specific to local rendering, then this should be specified in the thread.

[3rav]

~300MB .vtk:

just switch (without refreshing) and it works correctly:

[3rav]

Same behavior for pyvista 0.48 (NOT dev)

[3rav]

It's the same with the current libraries (this degradation still occurs):
trame 3.13.1
trame-vtk 2.11.8
vtk 9.6.1
trame-vuetify 3.2.2
trame-server 3.12.1

[bourdaisj]

Thanks for the follow up. Unfortunately as you might have noticed, we don't have a lot of time to look into it right now

[bourdaisj]

Hi @3rav ,
Following up on this - we have contributed a fix to trame-vtk in this MR that should fix the issue you are encountering.
Additionally, it seems that in your examples you were using/creating/destroying a lot of actors. In general, in VTK you should try to refrain from using too many actors.

[3rav]

My first tests show that the problem has been fixed, but I'll have to keep tinkering with it.

Thank you very much for your time and effort.

In the meantime, I switched to trame-vtklocal and it works even better, but I have problems with pyvista toolbars