search

guanyingc / SDPS-Net

Learning Based Uncalibrated Photometric Stereo for Non-Lambertian Surface (CVPR 2019 Oral)
http://guanyingc.github.io/SDPS-Net/
MIT License
178 stars 31 forks source link

Data capturing setup and number of light source #15

Closed zombicore closed 3 years ago

zombicore commented 3 years ago

hello 1- how does the setup looks like ? 2- how many light sources i need ? 3- can i make the light colored ? for example red blue green white ? 4- how many images do i need to take to make 3D object ?

guanyingc commented 3 years ago

Hi,

  1. For the data capturing, please refer to the DiLiGenT paper: https://sites.google.com/site/photometricstereodata/single?authuser=0.

  2. Normally our method requires around 8 light sources to produce robust results.

  3. This paper only consider white direction light source. You may search other related papers.

  4. Please note that our method only recover the surface normal map from a single view.

zombicore commented 3 years ago

Hi the URL doesnt let me read its content ( We are sorry, but you do not have access to this service. Please contact your Organization Administrator for access ) you mentioned "Please note that our method only recover the surface normal map from a single view." do you means that if i scanned the front view , it only show normals of one that view ? and if i need to make 360 i need to make other scans then do alignment ? does the system use CUDA for GPU processing ?

guanyingc commented 3 years ago

Hi,

  1. Please read "A Benchmark Dataset and Evaluation for Non-Lambertian and Uncalibrated Photometric Stereo, TPAMI2020" for a practical setup.

  2. Yes, if you scan the front view, our method only recover the normals for that view.

  3. This method can be ran in both CPU and GPU mode. https://github.com/guanyingc/SDPS-Net#q3-could-i-run-your-code-only-using-cpus

zombicore commented 3 years ago

Hi thank you a lot for fast reply i found that should use in the setup industrial camera , in the PDF they mention specific model , does that means the system already control the camera to take shots ? or/ i put the input images and it process the ply , kindly correct me

do i need Matlab software for this process ?

guanyingc commented 3 years ago

Hi,

  1. Please note that this code does not take care of the data capturing part.
  2. It can only recover the normals for image captured from a single viewpoint. You have to take care of multi-view processing and alignment by yourself.
zombicore commented 3 years ago

Hi one more thing i need to know , do i need Matlab software in any kind of processing ?

guanyingc commented 3 years ago

Using this code for normal estimation, we don't need Matlab for processing.