MVS for Emulated Perspective

[Clayton Pelz]

Dear OpenMVS Team,

I am an R&D engineer at simplehuman working on a new project I think should utilize MVS. After doing some reading on the subject, I would love to learn more and discuss my ideas with an expert. I will add some description about our project below to review, and if anyone in your team is interested, I can schedule a quick call. I am located in California, but I can be flexible with meeting timing depending on what works best for you. Please take a look and let me know:

Project description: 

We are making a product that would take a couple (maybe 3 or 4) images of the face from slightly different angles approximately 7" in front of a user's face. We then want to use those photos to simulate a new photo with a perspective 6' away from the face (standard portrait distance) and display that to the user. We think this would open up a lot of exciting possibilities for portrait taking.

Example input/output:

Considerations:

We can fully control the positioning and angle of the cameras on our device relative to each other and we can somewhat control the subject position in front of the device by automatically taking a capture based on subject position. The photos can also be captured all at the same time to eliminate issues caused by slight movement. Also, there can be lighting built into the device to make sure the subject is well-lit.

The ask:

I am mainly looking for an expert opinion on the feasibility of the project as well as what would be needed to get a proof of concept. Can we quickly set up a demo or is there some software out there we could reference that works similarly?

From there, I am also interested in discussing longer-term implementation depending on feasibility.

I am open to just having a quick and informal call or setting up some sort of consulting agreement.

Let me know if you have any initial comments or concerns.

Thank you for your time,

Clayton Pelz | Product Development Engineer, R&D | simplehuman

A: 19850 Magellan Drive, Torrance, CA 90502

P: (503) 894-4677

E: cp...@simplehuman.com

[cDc]

Thank you for your interest in our project. However I'm not sure I understand the project. If you want to obtain a portrait photo, why not taking one instead of the 4 around photos?

Anyway, OpenMVS can be use for this, and it will be a quite easy task given that the 4 photos you are taking are high quality and good lighting. However if you are interested in only the RGB photo and not the underlying surface as well, you could also use nerfstudio: https://github.com/nerfstudio-project/nerfstudio

I hope this is helpful.