Reference of neural network model for adaptation of 2D video for streaming to heterogeneous client end-points

1. A method of streaming immersive media, executable by a processor, the method comprising:

ingesting video content in a two-dimensional video format, wherein the video content comprises a first scene and a second scene, and wherein the two-dimensional video format references a first scene-specific neural network corresponding to the first scene, and a second scene-specific neural network corresponding to the second scene, the second scene-specific neural network being different from the first scene-specific neural network;

converting the ingested video content to a three-dimensional video format based on the first scene-specific neural network and the second scene-specific neural network; and

streaming the converted video content to a client end-point,

wherein the first scene-specific neural network and the second scene-specific neural network are referenced in a first supplemental enhancement information (SEI) message and a second SEI message, respectively, the first SEI message and the second SEI message included in a coded video bitstream corresponding to the ingested video content.

2. The method of claim 1 , wherein the converting of the ingested video content comprises:

inferring first depth information about the first scene using the first scene-specific neural network;

inferring second depth information about the second scene using the second scene-specific neural network; and

adapting the ingested video content into a first volumetric format associated with the first scene and a second volumetric format associated with the second scene.

3. The method of claim 1 , wherein the first scene-specific neural network is trained based on first priors corresponding to first objects within the first scene, and

wherein the second scene-specific neural network is trained based on second priors corresponding to second objects within the second scene.

4. The method of claim 1 , wherein a neural network model corresponding to at least one neural network from among the first scene-specific neural network and the second scene-specific neural network, and at least one parameter corresponding to the at least one neural network is embedded directly in at least one of the first SEI message and the second SEI message.

5. The method of claim 1 , wherein a location of a neural network model corresponding to at least one neural network from among the first scene-specific neural network and the second scene-specific neural network is signaled in at least one of the first SEI message and the second SEI message.

6. The method of claim 1 , wherein the client end-point includes one or more from among a television, a computer, a head-mounted display, a lenticular light field display, a holographic display, an augmented reality display, and a dense light field display.

7. The method of claim 1 , wherein the converted video content is immersive media that stimulates a somatosensory system of a user.

8. A device for streaming immersive media, the device comprising:

at least one memory configured to store program code; and

at least one processor configured to read the program code and operate as instructed by the program code, the program code including:

ingesting code configured to cause the at least one processor to ingest video content in a two-dimensional format, wherein the video content comprises a first scene and a second scene, and wherein the two-dimensional video format references a first scene-specific neural network corresponding to the first scene, and a second scene-specific neural network corresponding to the second scene, the second scene-specific neural network being different from the first scene-specific neural network;

converting code configured to cause the at least one processor to convert the ingested video content to a three-dimensional video format based on the first scene-specific neural network and the second scene-specific neural network; and

streaming code configured to cause the at least one processor to stream the converted video content to a client end-point,

wherein the first neural network and the second neural network are referenced in a first supplemental enhancement information (SEI) message and a second SEI message, respectively, the first SEI message and the second SEI message included in a coded video bitstream corresponding to the ingested video content.

9. The device of claim 8 , wherein the converting code comprises:

inferring code configured to cause the at least one processor to infer first depth information about the first scene using the first scene-specific neural network;

inferring second depth information about the second scene using the second scene-specific neural network; and

adapting code configured to cause the at least one processor to adapt the ingested video content into a first volumetric format associated with the first scene and a second volumetric format associated with the second scene.

10. The device of claim 8 , wherein the first scene-specific neural network is trained based on first priors corresponding to first objects within the first scene, and

wherein the second scene-specific neural network is trained based on second priors corresponding to second objects within the second scene.

11. The device of claim 8 , wherein a neural network model corresponding to at least one neural network from among the first scene-specific neural network and the second scene-specific neural network, and at least one parameter corresponding to the at least one neural network is embedded directly in at least one of the first SEI message and the second SEI message.

12. The device of claim 8 , wherein a location of a neural network model corresponding to at least one neural network from among the first scene-specific neural network and the second scene-specific neural network is signaled in at least one of the first SEI message and the second SEI message.

13. The device of claim 8 , wherein the client end-point includes one or more from among a television, a computer, a head-mounted display, a lenticular light field display, a holographic display, an augmented reality display, and a dense light field display.

14. A non-transitory computer-readable medium storing instructions, the instructions comprising: one or more instructions that, when executed by at least one processor of a device for streaming immersive media, cause the at least one processor to:

ingest video content in a two-dimensional video format, wherein the video content comprises a first scene and a second scene, and wherein the two-dimensional video format references a first scene-specific neural network corresponding to the first scene, and a second scene-specific neural network corresponding to the second scene, the second scene-specific neural network being different from the first scene-specific neural network;

convert the ingested video content to a three-dimensional format based on the first scene-specific neural network and the second scene-specific neural network; and

stream the converted content to a client end-point,

wherein the first scene-specific neural network and the second scene-specific neural network are referenced in a first supplemental enhancement information (SEI) message and a second SEI message, respectively, the first SEI message and the second SEI message included in a coded video bitstream corresponding to the ingested video content.

15. The non-transitory computer-readable medium of claim 14 , wherein the one or more instructions further cause the at least one processor to:

infer first depth information about the first scene using the first scene-specific neural network;

inferring second depth information about the second scene using the second scene-specific neural network; and

adapt the ingested video content into a first volumetric format associated with the first scene and a second volumetric format associated with the second scene.