Class Discussion Summary (Feb 18)

Put the citations to papers discussed here:

• Paper (from UCSC students) under review analyzing how good data needs to be to get a 3D Face

Scribes:

Christian Lei

Arik Yueh

Oasys Okubo

Esmaeil Mirvakili

Write the Class Discussion Summary Here:

Today you are practicing being a researcher. Instead of just understanding a past paper, we want to propose something new. Suppose we want to extend this paper from the synthetic examples where correspondence is known to "real" examples in which the input is actual image(s). Using images to get 3D is exactly like your HW goals. What would be our extension to the paper? What new piece of information could we add to the contribution of the paper? What plot/graph/figure/table/image would we use to support this new finding? Or alternately, what new experiment would better support the existing contribution?

You may find it useful to:

1. brainstorm lots of ideas,
2. pick 1 that your breakout group feels would be the best contribution.
3. design the ideal data/figure to support it. (This is exactly what we will be trying to do in part3 of the HW)

Checking for bias of methods with non-Caucasian faces, using varying skin tones may affect the error margins. (Arik)

We could use a combination of multiple inputs to lead to the outputs. Right now it seems that the input is only based on one input to see what angles would be required. (Christian)

Esmaeil:

We can print the ground truth model using a 3d printer. Then, we can use different cameras/scanners to get real images and data. In the end, the ground truth data is the same as the paper, so we can compare the result with the raw ground truth.