We are looking for a 3D computer vision engineer to extract facial feature locations from the scans that we capture from our 3D body scanner.
We capture full-body 3D scans with our scanner and have the ability to extract hundreds of measurements on the body, but recently it has become more and more important to be able to extract 3D location information for the following facial features as follows:
- the centroid locations of the eyes
- the centroid locationtip of the tip of the nose
- the centroid location of the chin
- the centroid location of the ear holes
- the centroid location of the mouth
- the widest points of the jaw (IE. the mandible joint)
It would be nice to have more information on each one of the aforementioned points, but the aforementioned is the bare minimum.
A few stipulations on the algorithm.
- An .obj file of the bust of a human
- A .json file containing the 3d location information for the aforementioned feature points as well as some confidence interval that this is an actual feature.
What will be provided:
- 100 bust scans (IE. the 3D .obj scan of a person from their bust up) as a control set in .obj format
- We will run the algorithm on the 100 scans provided as well as another blind sample set that was captured by the exact same scanning technology to evaluate the success of the algorithm.
- We assume that the algorithm will give 0 false positives and other than that, we'll have to evaluate with the contractor to evaluate acceptance of the project.
- Algorithm should be optimized for time and accuracy, whereby it should not take more than 5 seconds to process a single scan.
- The algorithm must not require any internet connection and all processing must be done local within the algorithm.
- Dependencies should be MIT license or similar
- Should run on CPU as opposed to GPU
- Code will be deployed on a cloud based linux environment
- Source code, test bench, build methodology, and executables