Poly Count

[gedsta]

Guys I have a few questions regarding poly count. This question arouse after I decided to help one of my coursemates with his models to render. His files are extremely large and come from a variety of programmes. his context is ~600k polys and his main building in the context is another ~200k polys. I think his buildings poly count is ok, but his context is just too big, when i merge both files it just lags tremendously.

 

I'm just curious how big your files become? How many you think its good for a decent interior render, how much is good for exterior (with trees and without)?

 

The file size and poly count, detail become an important topic especially when you work with not the best laptop, so you have to make the models efficient, cut out certain things and do them in photoshop.

[Corey Beaulieu]

If you want great renders, poly count should not be a concern of yours. I don't mean this in elitist terms, but that your computer will lag when poly count is high so you need to learn to work with it. Isolate everything when you work on it and be smart with your layer management. Proxy when you can and when it is reasonable and display as box the things that are complete.

 

We frequent a heavily proxied model that includes forest pack that still reads a poly count of 10-30 million polys. This is due to our scale of project, but when I get a more basic interior it makes me want to model everything as it only means that I can go nuts on details.

 

I do no believe that anyone should be concerned with poly count until their machine breaks and only then should they consider the opulence of too much detail.

 

Save selected and merge back into things. Don't resort to old methods of trickery.

[marius e]

That is just bad advice Gedsta. Polycount should be as low as possible. The more complex ones work get the more poly's you will have yes but any respectible 3d artist will keep it as low as possible. Good modeling = Low poly at visibly good standard. Lazyiness and bad modeling = High polys in scenes where one does not need it. MOST works can be low poly counts, not everyone has city scapes where you have millions of polys, that is a different issue.

 

That is probably the worst advice I have ever seen Corey Beaulieu. Not only wasting your own time but wasting company time and project time.

[gedsta]

Well, it is not game design so the poly count has a different meaning in architectural rendering. Corey Beaulieu did have good advice in there, any advice is welcome, I just want to have a conversation here. When I do my own works in 3ds max the poly count is low because I model most of my building in basic blocks and add detail only where the cameras are going to be set up and i try to take into consideration the distance of the camera from geometry, so the poly count always goes up to 500k at most with trees and furniture (my designs are not a very huge scale also).

 

The problem with the poly count/file size is when you get files from another person who is not as good in modelling and just makes a huge model with a lot of unnecessary detail (sketchup model with a lot of 3ds warehouse stuff just put in there). How would you go about fixing up a file like this? (take note its hard to navigate and select things). I did ask him to redo certain things and reduce the file size, but part of me just wants to give up on helping him because it is a slight pain.

 

Which brings me to another set of questions: Clients...Is this what an Arch visualizer has to deal with, do you usually get revit files or semi finished files from your clients and just redo the materials and render? or do you specifically ask for plans/sections/elevations and remodel everything from scratch?

[Scott Schroeder]

I don't think what Corey said is the worst advice in the history of CGArchitect. However, you do need to be concerned with how you model things and not have 6 bazillion polygons on a rock that takes up .05% of your scene.

 

Not seeing an image of the scene, it's hard to tell if the 800,00 polygons are worth it or if there is area for improvement. But off the top of my head I would advise you to start saving for a new computer. 800,000 polygon scenes are pretty normal these days. Thankfully it's not 1980 anymore and we don't have to overly fret about if we can afford to make a cube out of 6 polygons or if we need to optimize it further.

 

There is a very thin line between being overly paranoid about your polygon count and just modeling like you don't have a care in the world. Both theories will cost you production time and will give you poor results in the end. You have to find a balance between then two and usually that balance is dictated by two things. The hardware you are on and your desired level of quality. The higher the quality the more polygons you are going to need (unless you want to do a lot of high to low poly baking). The more polygons you have, the better hardware you need to display them.

 

An exterior view with 3d trees, I would expect to come in well over 1 million polygons and I wouldn't be shocked to see an exterior with heavy landscaping to be 30 million or more polygons. Most really good 3d tree models are millions of polygons in themselves. As Corey said, most major render engines support proxy objects so landscape polygon count isn't as limiting as it used to be.

[heni30]

I'm a fanatic of keeping things lean. I just start getting very frustrated when a viewport rotation starts stuttering or autobackups start taking too long, which sours things for the duration of the project. I'd rather spend time making a scene efficient in the beginning than figuring out strategies to handle a bloated scene.

Edited May 24, 2015 by heni30

[Corey Beaulieu]

To clarify my opinion a bit, I don't suggest that one model with abandon. I just have grown tired of bump/normal mapping and calling that detail. In hi-res renderings it really stands out to me. LOD's are obvious wise, but here too I find it excessively time consuming to the work I do because I am often rendering the same space from many angles and what was once in the BG is now front and center.

 

Many models come from libraries now days and the better ones have a lot of polygons. You would be a fool to spend too much time optimizing them. I think you should be mindful of their need for detail and then take the fact that someone saved you a lot of time by doing a good job there and spend your time modelling more detail of the architecture. Floor geometry for instance is one I almost always model (for a lot of reasons), but I still see it being mapped onto a plane. I think we can do better.

 

Poly count is not a concern for me because I do have a decent machine, a high level of detail is not overly outrageous poly-wise when well considered, and if it is our goal to be physically accurate, we should model geometry so we can better map it. Lighting then appears more natural because it's bouncing and reflecting more appropriately.

 

This will cost you at render time, but not insanely so. It will be more rewarding in my opinion. Render time is not artist time in my world so I have a low level of concern for it. I only seek to keep it manageable.

 

My advice was and still is to learn to work with heavier scenes. They are common in large scale work.

 

As to your second question, we receive a lot of Revit, Rhino , and Sketchup models. They never transfer well so much clean up needs to be done before materials, lighting, and rendering can happen. They are extremely slow at first due to high OBJECT counts, not poly, and the work is largely to collapse them into common types. Most of the time CAD supplements help to increase model detail as they are often not modeled by the architect, but you never know what you are going to get.

 

The overarching advice from everyone here, whether they agree with me or not, is to work smart. Plan out your project for your end of it. Your client can't be overly helpful in trimming your model for you so you have to learn to work smart with what you are given and for what you are trying to do.

[gedsta]

Solid advice. Best thing for me to do at the moment is to save up for a good stationary computer with dual or three monitors after my student days are over. I think I just get frustrated like hell when the computer keeps lagging/freezing, takes all the joy out of modelling/texturing.

 

Shouldnt Revit import nicely into 3ds max? I tried that a few times with my projects and only saw problems where I modeled in Revit poorly.

[Corey Beaulieu]

Smaller models import rather well and truthfully, large ones would too, but as you said, when they are modeled poorly, they import poorly.

 

Large models tend to have the least amount of organization. I've had a few good ones, but mostly they are made by too many hands and the automated stuff goes haywire. Gneric materials screw everything up.

 

2015 layer manager proves it's worth when you have to do an old school transfer. Selecting through it and organizing by alphabet makes a lot of lag go away. I was able to, just this past week, convert models of 20-50k objects down to 200 or so in just over an hour each. I'll take that as a win.