Sky background model
An AR client device generates and uses a background model to identify portions of images that depict the sky. A background model is a model that represents where the sky is visible for the client device. To identify a sky background portion of an image, a client device can map an image onto the background model and thereby determine which portion of the image represents the sky. The client device can use the identified sky background portion to augment the image to include AR content in the sky. To generate the background model, the client device applies a background detection model to a set of images to generate background probability images. The background probability images are mapped onto a background model using orientation data captured by the client device to update the background model based on the background probability image.
1 . A computer-implemented method comprising:
accessing an image captured by a client device, wherein the image depicts an environment around the client device and wherein the image comprises a sky background portion and a non-sky background portion;
accessing orientation data describing an orientation of the client device when the accessed image was captured by the client device;
generating a background probability image for the accessed image by applying a background detection model to the accessed image, wherein the background detection model is a machine-learning model trained to identify pixels in an image that correspond to a sky background portion of the image;
mapping pixels of the background probability image to portions of a background model based on the orientation data, where the background model is a model that indicates where sky is visible by the client device;
updating portions of the background model based on the mapping of the background probability image; and
storing the updated portions of the background model at the client device.
2 . The method of claim 1 , wherein the image is a frame of a video captured by the client device.
3 . The method of claim 1 , further comprising:
establishing an application session with an online system; and
responsive to establishing the application session, initializing the background model.
4 . The method of claim 1 , wherein each pixel in the background probability image corresponds to a pixel in the accessed image, and where each pixel of the background probability image indicates whether the corresponding pixel in the accessed image depicts the sky.
5 . The method of claim 4 , wherein a subset of pixels of the background probability image are labeled as depicting the sky.
6 . The method of claim 4 , wherein each pixel of the background probability image comprises a likelihood that the corresponding pixel in the accessed image depicts the sky.
7 . The method of claim 1 , wherein mapping pixels of the background probability image comprises:
determining a field of view of the accessed image based on the orientation data.
8 . The method of claim 1 , wherein the background model comprises a 3D structure that is located a set distance from the client device.
9 . The method of claim 1 , wherein the background model comprises a rectangular prism.
10 . The method of claim 1 , wherein updating portions of the background model comprises:
updating probability distributions associated with the portions of the background model.
11 . A non-transitory computer-readable medium storing instructions that, when executed by a computing system, causes the computing system to perform operations comprising:
accessing an image captured by a client device, wherein the image depicts an environment around the client device and wherein the image comprises a sky background portion and a non-sky background portion;
accessing orientation data describing an orientation of the client device when the accessed image was captured by the client device;
generating a background probability image for the accessed image by applying a background detection model to the accessed image, wherein the background detection model is a machine-learning model trained to identify pixels in an image that correspond to a sky background portion of the image;
mapping pixels of the background probability image to portions of a background model based on the orientation data, where the background model is a model that indicates where sky is visible by the client device;
updating portions of the background model based on the mapping of the background probability image; and
storing the updated portions of the background model at the client device.
12 . The computer-readable medium of claim 11 , wherein the image is a frame of a video captured by the client device.
13 . The computer-readable medium of claim 11 , the operations further comprising:
establishing an application session with an online system; and
responsive to establishing the application session, initializing the background model.
14 . The computer-readable medium of claim 11 , wherein each pixel in the background probability image corresponds to a pixel in the accessed image, and where each pixel of the background probability image indicates whether the corresponding pixel in the accessed image depicts the sky.
15 . The computer-readable medium of claim 14 , wherein a subset of pixels of the background probability image are labeled as depicting the sky.
16 . The computer-readable medium of claim 14 , wherein each pixel of the background probability image comprises a likelihood that the corresponding pixel in the accessed image depicts the sky.
17 . The computer-readable medium of claim 11 , wherein mapping pixels of the background probability image comprises:
determining a field of view of the accessed image based on the orientation data.
18 . The computer-readable medium of claim 11 , wherein the background model comprises a 3D structure that is located a set distance from the client device.
19 . The computer-readable medium of claim 11 , wherein the background model comprises a rectangular prism.
20 . The computer-readable medium of claim 11 , wherein updating portions of the background model comprises:
updating probability distributions associated with the portions of the background model.