Open AIMads opened 2 years ago
Hi, when trained on ScanNet dataset, the depth shift is 1000. So the per-pixel depth value is in "mm". However, so many aspects limit the depth prediction from a simple monocular estimation model to be generalized to new camera intrinsic and new scenes. I am afraid there is no simple way to get any usable depth estimation in meters for accurate floor planning. In fact, to make the depth estimation consistant between frames is already challenging, if not trained with a specific loss function.
On Mon, 22 Aug 2022 at 11:45, Mads @.***> wrote:
Im interested in seeing if you can use this method for performing floor planning and was considering if there was a way to extract depth estimation in meters from the model prediction?
— Reply to this email directly, view it on GitHub https://github.com/EryiXie/PlaneRecNet/issues/15, or unsubscribe https://github.com/notifications/unsubscribe-auth/AG7TXSZD3W36U5L7EOZP3SLV2ND2LANCNFSM57G7GUZQ . You are receiving this because you are subscribed to this thread.Message ID: @.***>
Okay dammit! But thanks for the detailed answer!
Im interested in seeing if you can use this method for performing floor planning and was considering if there was a way to extract depth estimation in meters from the model prediction?