0
\$\begingroup\$

I have a question regarding to the HDRI maps as many some of you have a bit of experience in this field especially for Oculus Rift platform!

I've been trying to achieve a 3D landscape from this HDRI map. (Also I highly recommend this website for others ​:)!)

I even decided to render in Maya few images from the orthographic perspective but unfortunately still no luck.

See the results here on YouTube.

Is there any other method to reconstruct 3D models just from images and use them to create environment?

\$\endgroup\$
3
  • 1
    \$\begingroup\$ One thing to watch out for is that the HDRI map you linked includes specular reflections (the glare off the pavement, the reflections in the windows). These reflections are only correct from the viewpoint the image was taken from. They won't automatically move to follow tracking of the head, or even the parallax between the player's eyes, which can make them look flat and painted-on. \$\endgroup\$
    – DMGregory
    Mar 16, 2018 at 1:27
  • \$\begingroup\$ DMGregory thank you for reply. I believe that I need to 3D model every detail then. Is there any way to achieve photorealistic effects using just Photographs? Also I want to share with you the idea which I am going after regarding the environment: youtube.com/watch?v=BT0X9ZoyaDY&t=6s \$\endgroup\$ Mar 16, 2018 at 8:10
  • \$\begingroup\$ Photographs, plural, yes. There are systems that will cross-reference many views of an object/scene from different viewpoints to try to separate what portions of each pixel are due to material albedo, diffuse lighting off a surface with particular orientation, or specularity. But what you're providing as input here is a single photo, with nothing else to cross-reference against. Inferring all that layered information from one view is still the realm of state of the art AI. Or you could hire an artist to build the scene to spec. \$\endgroup\$
    – DMGregory
    Mar 16, 2018 at 9:58

1 Answer 1

2
\$\begingroup\$

It's very hard to reproduce a 3d environment from a single image, in fact, it's pretty much impossible, even for humans.

For instance, these images:

enter image description here

You can't really tell whether the man at the back is a giant or they are simply the same distance away from the camera.

However, if you have a lot of images with known positions of the same area, then you can reconstruct it. This is called photogrammetry.

It's usually done using preprocessing (because it can take a long time), but real time versions do exist.

\$\endgroup\$
1
  • \$\begingroup\$ Sure Balint and is that also applies to buildings? What if I want to create buildings somewhere far in the background and fake it with polyplanes? I guess photogrammetry method will be the best option here! Also I want to show you what I really want to achieve. Here is the link for the video: youtube.com/watch?v=BT0X9ZoyaDY&t=6s \$\endgroup\$ Mar 16, 2018 at 8:06

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .