I'm attempting to build a model of a car fender. The goal is to modify this in software, CNC route a mold and fabricate the resulting part.

The original part is fairly complex, with a bunch of compound curves and things that look perpendicular but aren't quite. I do not have access to a 3D scanner. So, I thought I'd try AutoDesk ReCap Photo to generate a model, clean it up and go from there. This is what I got from ReCap:

enter image description here

enter image description here

Getting rid of the background trash, etc. was straightforward. What I'm trying to do now is clean up the mesh itself, where there are a lot of bumps (see the bottom of the fender in the image, and areas around the vent towards the top of the piece).

I'm currently waiting on ReCap to generate a new model from much higher-res images of a much cleaner part. That will hopefully reduce the amount of noise in the model.

The Question: What techniques should I consider to smooth the surface to get rid of bumps, dips, and generally remove noise from the mesh, or should I be using a different technique to generate the model?

I considered creating a lower-res quad patch and using the Cloth modifier to drop it onto the fender. That (perhaps obviously) didn't work very well. :/

  • Welcome to GD.SE - Please look through tour to get a sense of our community – who we are and what we’re about. Then look over How to Ask and How to Answer a question to see what makes a good query here, and how best to frame it. Ironically, although I often champion including 3D generalist questions in our scope, this one seems overly 3D specific: you're looking for photogrammetry cleanups & retopology workflows in 3DS & I I'm really not sure that fits our scope. Please read read through How to Ask and if you can, edit your question to better fit our scope. Thanks! Oct 17, 2019 at 20:36
  • Fred Flintstone would be quite happy with your fender. But your surface accuracy is useless when compared to how precise car parts should be today. In the passed away Soviet Union your quality might have been acceptable at least when sold to ordinary people who have stayed in the queue 5 years, but now you should have better model. Smoothing will not make it if you have no way to quarantee the smoothing pulls the dimensions right, not away from the right. Did you have any kind of high contrast non-glossy "make it uniquely traceable" surface texture during the scan?
    – user82991
    Oct 17, 2019 at 21:54
  • @user287001 dimensions are fine. I realize that the scan is rough. No, there were no surface texture / mapping / reference points. These are the things I'm looking for in an answer. I realize the model is crap - I'm looking for a way to either clean this one up, or a way to generate a model that is NOT crap. This is for my track car - it doesn't have to be perfect, just better than this mess.
    – 3Dave
    Oct 17, 2019 at 22:20
  • 1
    @GerardFalla Apologies if this is not the correct SE. I asked here because there are more modeling & MAX questions here than other sites in the network. I'd appreciate a pointer to the correct site, if you have any suggestions.
    – 3Dave
    Oct 17, 2019 at 22:21

1 Answer 1


Photogrammetry is statistical estimation which tries to find the most probable reason for the given set of photos. That includes reasoning what's the object, where the camera has been and what error the camera causes.

Photogrammetry can work only if the light is stable during the session, the surface isn't glossy (=all reflections are diffuse) and there's enough unique surface details with high enough contrast. As important is to prevent all changes in the camera between the shots. If moving the camera affects the light in the room or the camera does automatic adjustments or intelligent postprocesing to make the image better, you are out of luck.

The surface must not be glossy and it must have high contrast details. Stone and wood are often ok as is, but metal, plastic, glass and painted surfaces generally must be prepared. A common way is to have matt water based grey paint and a black fine, but still well visible pattern on it. If there's glossy areas, they appear as holes or unexpected bumps.Lack of details causes the same.

Having a dense high contrast pattern doesn't prevent surface errors because photo pixels still makes everything a little random. But the pattern can prevent large bumps, so it's more probable to get usable cloud decimation result.

It's far too easy to forget the near environment. It must also be traceable to prevent mixing it with the target.

Proper (=diffuse shadowless and stable) light is very difficult task. I have seen tutorials where the beginner is guided to shoot outdoors in a cloudy day and to work fast to have consistent light. It worked for me, too.

Search for quidance how to prepare the object and how to shoot the photos. Start from this: https://www.3dscanstore.com/blog/3d-scanning-reflective-objects.

  • THANK YOU. I'm a graphics coder and a car guy; not an artist or capture expert, so there's a lot going on here that is outside of my expertise. Based on your input and the link, I have a pretty good idea of what my next attempt will entail. Thanks so much for taking the time.
    – 3Dave
    Oct 18, 2019 at 16:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.