I’m new to reality capture and need a little help and advice.
I’m trying to create a full 360 textured model of a vintage Nikon camera.
I’m having a problem with the fact that the camera has both a nice textured surface and also some smooth reflective surfaces ! I’ve attached a couple of data set images and also some screen shots of the generated model.
I completed 6 rotations of the camera, each rotation with the object sitting on a different axis (roughly 50 images per rotation) and processed in lightroom (reducing highlights and increasing shadows) before inputting to RC.
I then processed each pass in RC separately and created a masked from the preview reconstruction.
I aligned all components in a new project with some manual control points, and diasbled a couple of images that wouldn’t align.
The mesh is very noisy, although with a good amount of detail but the gloss(flat) parts of the camera are destroying the model .
when photographing shiny objects, it’s important to help the algorithm as much as possible. One option is to use this workflow How to reconstruct shiny objects with RealityCapture , but in your case you probably do not want to spray mat paint all over your object. These surfaces are not only shiny, but also featureless, making it very hard to triangulate any points. It’s important to consider the specular highlights on these mid-sheen to high sheen surfaces. For instance when that large window is behind the subject, There will be very high specularity on the surface, as illustrated here
which will massively drop as you move around, causing drastic changes on these surfaces from input to input. This is a situation where a turntable shot from the window side would really help. Your next best option would be to shoot in such a way that the window or any strong large light source is not directly behind the subject, maybe in the middle of the room on the floor. Also if you are shooting in RAW it can really help to pre-process your images, and try to level out the shadows and highlights, but don’t go too extreme,something like this, except tweaked for the raws
Just make them a little flatter. The other thing is, your first alignment will go a bit better if you can cover any shiny surfaces (like the marble top) around the subject in newspaper, but make sure it is firmly stuck down with tape, so it can’t move.
Thank you for your response, lots of helpful info for me to digest and implement.
I was struggling to get my iso down to 100 while maintaining an aperture greater than f8, hence the large window in the background, but your advice regarding specular highlights makes perfect sense, so will adjust my setup positioning.
I’ve ordered a cheap, flash mounted diffused led light panel. So that the object will always be front lit with a soft light source. Do you think this will help ?
I picked the camera as my first attempt at photogrammetry, as I knew it would be challenging, having both matt and reflective surfaces and complex geometry. I possibly should have started with something simpler !! However, I want to be able to produce a consistent process and workflow that I can replicate, so started in the deep end.
I like the idea of the turntable but have some large and heavy objects that I wish to do next, so maybe not practical for me.
I do shoot in RAW and pre-process in Lightroom, but maybe aggressively not enough. When I re-shoot i’ll stick to your advice and tone curve above.
When exporting, i’m doing so as 24mpx jpegs at 300dpi. Would you recommend exporting to a different file type for use in RC ? Will it make a real difference ?
100 ISO is ideal, but you can check how noisy your camera gets at higher sensitivity, you might find you can use a bit more.
I think the LED flash should really help, but obviously be aware of the nasty flash shadows.
That’s certainly an ambitious first project, but I think that is a great approach as it’s going to be so easy on mat objects with plenty of features.
My curve was done very crudely, it may be that your pre-processing was better, I just wanted to show a rough example.
I think those JPGs will be fine. If you have a true 16bit workflow you can use tiffs, but they are only really useful for texturing, not for alignment and meshing, but as you suggest not a profound difference, especially on a model like this.