NeRF - Neural Radiance Fields - Testing #0T007BD0C

NeRF - Neural Radiance Fields ???

?? Prerequisite (Hopefully you have read my article on NeRF first): https://www.dhirubhai.net/pulse/nerf-neural-radiance-fields-3d-scanning-0s007bc0h-vijay-parmar/

I had hands on this NeRF technology and found it very promising for creating high accuracy 3D models of real world compared to photogrammetry technology.

So, Let me explain this technology in very layman language:

You give some of the surrounding photographs to the

Following is my own 3D made out of 1590 photographs.

So, how i did it ?

1??

1. Capture/Scan Data with help of good AR supported camera or Drone [For good final results - Use camera which takes photograph with EXIF data which contains GPS location, Time, Camera Orientation and other camera details like focal length and all etc.]

2??

2. You feed the data into NeRF Tool like Nerf.jl or Luma AI, The tool will imagine complete 360 scene out of your given data; the more data you give, the more accuracy you will get.

So assume that i have only 3 Photos with me (120° each),

NeRF will try to imagine 360° angle in all the axis from given data with help of AI, and the similar like photogrammetry (point cloud) tries to combine images to make it 3D but with different approach.

3??

3. So just like long exposure photography, camera captures light for very long time with open shutter; which helps it capture dim-lights. So, in very simple language, the light which stays longer in frame or very high nits light goes to frame stays in frame !

NeRF tech works like kind of same style, it looks in to all the frames and draws rough sketch in all possible directions and makes lines darker and darker which matches to other frames.

4??

4. The model gets fine tuned in some time and ready for use. We cab use it for primary reference file to model over it, or can use it to get volume or can have look of object in 3D AR/VR glasses.

So Quick use of Civil Engineering are:

1. Get volume of material on site (like sand, or aggregate from its stock)

2. SCAN2BIM - Create BIM model over 3D mesh made from NeRF model [Just like we do over point cloud]

3. Structural inspection : Not useful [As it imagines missing data in-between, we can only rely on NeRF if enough data captures]

I am exploring its use in structural engineering so, Stay Tuned till i post update in use of this NeRF technology in our Structural Engineering.

Some of my other Experiments:

Links:

• Nerf.jl : https://github.com/JuliaNeuralGraphics/Nerf.jl
• 英伟达 Blog: https://blogs.nvidia.com/blog/instant-nerf-research-3d-ai/
• Luma AI : https://www.luma-ai.com/blog/
• Tooliqa Article: https://www.tooli.qa/insights/neural-radiance-fields-nerf-a-breakthrough-in-3d-reconstruction
• Good Research paper by Kyle Gao on NeRF : https://arxiv.org/pdf/2210.00379.pdf