Closed AB00k closed 3 months ago
Hi, In visibility.txt you get binary values - 0 or 1, which indicate whether a particular landmark point is visible or not. 0 indicates visible and 1 indicates not visible. Below is the mapping of those 29 landmark points.
In attributes.txt, you get binary values for the attributes. 1 indicates that a particular attribute is present in the face. The ordering of attributes is same as in CelebA.
But the response that I'm receiving has 29 values in visibility.txt and also I wanna know if there is any relation between landmarks and visibility output from model that we can use like if I want to see which landmark point represents which point in visibility.txt is it possible.
Hi, The visibility output predicts the visibility for 29 landmark points, however we don't predict the coordinates for those landmark points but only the visibility. We use the % of visible landmark points as the estimate of the visibility.
We don't provide a direct prediction of landmark and its visibility. However, if you want the exact mapping of landmark and its visibility, I suggest you to convert the 68 landmarks that are predicted to 29 landmarks using 68 --> 29 landmark mapping, which might be available online (not sure).
I'm uncertain about the meaning of each value in visibility.txt and attributes.txt. I'm particularly interested in understanding visibility.txt, as I suspect it's related to landmarks, but I'm unsure how.