Jump to content


  • Content count

  • Donations

    0.00 CAD 
  • Joined

  • Last visited

Community Reputation

1 Neutral

About Adrian

  • Rank

Personal Information

  • Name

Recent Profile Visitors

1,595 profile views
  1. Andy58, Not using Single Scattering made the key difference. Without it, the material lost a lot of definition/contrast, but with more tweaking I think I'll be able to get a satisfying result. Thanks a lot for your help!
  2. Juraj, something like that, yeah. Arnold looks really cool btw, very intuitive SSS parameters! Andy58, unless I made some other mistake, I don't think you suggestion works. I tried the "Full Ray Traced" model (though that would be too slow for my needs anyway), but it shows identical artifacts to the "Global Point Cloud" (fast). I've attached two images, only differing by the amount of attenuation set in the shader. The scene is lit from the front by a point light, as you can see by the shadow thrown from the top cylinder onto the one behind/right to it. However there are also dark see-through like shapes that wouldn't be there in a unified volume; there should be no cylinder going though that box anymore. The artifacts don't behave like shadows or reflections, which I suspected at first. Even though the second image looks much better in this regard, it essentially suffers from the same problem, just masking it out more. p.s. Sorry for not aligning the images, the forum software doesn't seem to allow me to specify that.
  3. Hello. I was wondering – can anyone think of a way to compute a SSS point cloud that treats closed, overlapping geometry as a single volume. I have meshes that would be very tedious to unify, and I doubt poly -> VDB -> union -> poly -> reduce can give me a crisp enough result. Alas I don't have enough insight into the workings of how SSS point clouds are mapped onto polygonal geometry, but wondered if a volumetric approximation of the meshes with sufficient resolution could serve as the basis (somehow). Any ideas / suggestions welcome.
  4. Thanks for the suggestions Dennis! I only display the geometry as a point cloud in the viewport - the connecting primitives outnumber them about 60:1. I tried the PolySoup SOP but like prior attempts, mantra seems to find no way around having to see all geometry at once (at some point). I'll try to find a way to create a new point each time my PythonSOP tries to make a new connection. If I retain the original point cloud and hide the newly created ones, I imagine I can still work in the viewport; the locations stay the same after all.
  5. The convert-line SOP looks interesting, but it appears that it does nothing to a structure such as the one in my screenshot. I currently create the "network" by first placing points and then running a Python SOP to create polygon primitives connecting various points (using external data). In my thinking, the shared points are what makes the structure indivisible. I was already using the micropoly renderer and a flat color shader, but I haven't tried rendering the screen in multiple segments, that might actually work! Thanks! I'm always so worried about drive longevity, a bad habit I guess. I'll consider it as a last resort, if only to see how the end result would look. I have 64GB RAM, which is the maximum supported by my CPU/Mobo, and the geometry data itself fits into that - it's the renderer's acceleration structures which make this balloon beyond capacity I think.
  6. I'm trying to render a very large collection of points connected by lines (twin-vertex open polygons) which in Mantra unfortunately exceeds the amount of memory I have available on my system. This is an attempt at data visualization of node connections, so correct representation of the data matters to me, even if the viewer couldn't possibly notice. It's similar to the image below, just orders of magnitude larger. I made sure to get rid of any geometry attributes not needed for rendering. Since many points are shared by different lines (i.e. the whole thing is interconnected), I don't see how I could e.g. partition the network into cubes of space to help memory management. To my understanding, none of the nifty tools like packed primitives can help me when dealing with one contiguous piece of geometry. I played around a bit with mantra's KD-Tree settings but it's not enough to save me. If it just isn't doable, I'll have to scale back my ambitions, but right now I'm still hoping to find a way. Any ideas/suggestions welcome! Thanks.
  7. Edit: Solved, input attribute needs to be of type float to be interpreted correctly, int is not ok apparently, even if converted. I have a bunch of points that carry a "refCount" point attribute of type integer. I wanted to create a simple flat shader that colors geometry based on this attribute. I had a vague memory of how to accomplish this and looked in the forums to confirm this. However it appears that the approach I tired doesn't work; in mantra, the points only carry the leftmost color of the ramp, i.e. the one closest to 0. I've attached a screenshot of the network as it's simpler than explaining. The fit range node takes in the expected range of "refCount" and outputs values from 0 to 1 for the Color Ramp. I have enabled attribute visualization in the viewport and can see that Houdini picks up the "refCount" attribute just fine. What am I doing wrong in the shader?
  8. I had the same issue while briefly test-driving GNOME3, so I think the problem is not XFCE specific, and H14 worked fine as I recall. The XOrg logs show no problems either. One of the first things I did was to disable XFCE's compositor, but again no difference. I can't think of what to do next to narrow down the source of the problem. I'd rather not try newer graphics drivers, as that might introduce problems of its own. UPDATE: It appears to work now. Sorry, this is a little embarrassing, but I just fired up my workstation again to investigate further and the problem appears to have vanished entirely. As i write this, I'm doing a heavy render using all threads, without even touching process priority, and experience no lag at all. The preview render is completely responsive, it's like night and day. So unless it comes back (knock on wood), don't waste your time and consider my thread void. Bad start into the new year averted
  9. Hello, I'm having trouble with Houdini 15 on my Xubuntu machine. At first I thought it was a problem with the X-Server, because open windows wouldn't redraw themselves and I couldn't open any new ones, but then I realized that setting the mantra threads to 1 seemed to alleviate the problem (8, 7, 6, were all equally bad). It appears that spawned mantra instances kept rendering in the background, which I failed to notice from the unresponsive UI. So now my guess is that mantra hijacks too many system resources, completely stalling the Desktop environment. This applies to the Houdini UI itself too by the way - a few seconds into a preview render, the image doesn't update any more. The same with MPlay when doing an interactive render. I tried starting up Houdini with a "niceness" value of 19 and an mantra instances inherit that, but it doesn't make one bit of difference. I have a fairly beefy system, though with a consumer graphics card (Nvidia GeForce GTX 970). Nvidia drivers are the most recent ones available in the regular Ubuntu repository, i.e. 352. Do you guys have any suggestions on what might alleviate the issue, or maybe have experienced it yourself? I vaguely recall that setting priority low worked for me in the past, but that was back on Windows I think. Thanks.
  10. Smart man! That could well be it. Someone at SESI took a short for an int, eh? Edit: I think you were right on the money. I just tested 32000 vs 33000 and that's where the crash happens. I also rendered a few duplicates at 32k without a crash. Houdini 12.1.125.
  11. I'm having issues rendering the object below. Starting in the 40-50K points ballpark, i get a memory allocation error when starting a render with Mantra. As you can see, this object is very light weight, so i can't see this being limited by mesh density this early on. Here a few tests: file node with bgeo source: mem error when reaching certain point count (see above) file node appended by convert-SOP of type nurbs curve: mem error (same threshold) rendering geometry node as points only: works, in the millions even. file node appended by polywire-SOP: works! way denser mesh! Polywire is neat, but extremely slow and i'd actually rather render curves with a constant shader so as not to consume too much memory. Any ideas what could cause this?
  12. Thanks for the answer and files rafaelfs, i will have a look at them. I'm a bit too eager and impatient sometimes, this will take time to learn. The reason I was interested in running python from the code tab was due to what appears to be a restriction/limitation of the script tab regarding geometry manipulation. It seems code within the code tab has more permissions, unless i am mistaken.
  13. I'm trying to create a python asset (or "operator type" or whatever these things are called). I read that geometry is read-only, unless accessed from code contained in the code tab. Here are two problems i can't get past: 1) I would like to create a parameter interface and have value changes in it call a function in the code tab. The callback field however only seems to recognize functions from the script tab, so no luck there. Can i get to code somehow, or is this in conflict with Houdini's node paradigm? ? 2) The kwargs arguments passed by the callback field only seem to apply to the parameter-item that called them. Can i make a button trigger a function, but pass the value of e.g. an integer field to that function, or alternatively retrieve the value from within the function without it having been passed?
  14. The add SOP worked perfectly for what i wanted to achieve, thank you.
  15. Hello, I'm new to this forum and fairly new to Houdini. I have two questions that i couldn't find answers to via search engines or the help files. 1) Looking at the picture below, what type of object are the black lines connecting the points? By that i mean, do they have a name and what is their type within houdini (e.g. primitive, polygon, edge)? 2) I want to write such points, connected by lines, using python. For a sequentially connected series as shown in the image, the curve SOP seems to work well (and draws lines automatically). Let's say i wanted to connect the points more arbitrarily, e.g. single points connecting to multiple others; is that possible with these (point/line) objects, and if so, into which type of SOP do i write them? Thank you, Adrian