It is as good as I believe can be done with the current limited data set of images that have been publicly released by the Rosetta mission. It was built by correlating 150 surface features over 120 NavCam and OSIRIS frames. Each image has had its viewing geometry reconstructed using sparse bundle adjustment (according to the technique explained in this blog post). Then, a select number of images were used to derive dense depth-from-stereo data. To combat the noise in the stereo data, I used shape-from-shading to create high-resolution local data.
For the parts of the comet currently in darkness I have used images of the limb against the slightly brighter dust cloud to constrain the volume that the surface is inside.
I have smoothed out the “unknown” areas to make the model a bit prettier.
There are areas with artifacts and seams, especially around the neck. I have not fixed them because the model is continuously made obsolete by the steady stream of new images being released.
As soon as new data becomes available I try to update the model. And I will post a new one here if there are meaningful changes to the model.