Open hua-wu-que opened 3 years ago
Same question.
And what is the role of focal in the algorithm? Can BTS be applied for videos without camera intrisincs? How does BTS address the global scale ambiguity as in many monocular depth estimation methods?
Hi,
I observe some code repo "monodepth2" for example, has median scaling turned on during evaluation. (multiply the predicted depth by a ratio compared to gt). There is no such step in this repo and the performance is still much higher?