LATENCY MANAGEMENT WITH DEEP LEARNING BASED PREDICTION IN GAMING APPLICATIONS

1 . A method for reducing a latency in an interactive application comprising:

obtaining a first frame, the first frame being representative of a first action performed by a user in the interactive application;

obtaining information representative of a second action performed by the user in the interactive application, the second action following the first action; and,

predicting a second frame corresponding to the second action from data comprising at least the first frame and the information representative of a second action using a neural network.

2 . The method according to claim 1 wherein the method further comprises displaying the second frame.

3 . The method according to claim 1 wherein the method further comprises obtaining metadata along with the first frame, the metadata being at least representative of a status of the interactive application at a time corresponding to the first action and/or of the first action, the second frame being further predicted from the metadata using the neural network.

4 . The method according to claim 3 wherein the metadata representative of a status of the interactive application comprise information representative of the user and/or information representative of dynamic objects and/or of other users in the interactive.

5 . The method according to claim 1 wherein the neural network use parameters:

trained offline using data representative of frames, user actions and status of the interactive application collected during an offline execution of the interactive application; or,

trained on the fly using data representative of frames, user actions and status of the interactive application collected during a current execution of the interactive application; or,

initialized at a start of an execution of the interactive application using parameters trained offline using data representative of frames, user actions and status of the interactive application collected during an offline execution of the interactive application and then trained on the fly using data representative of frames, user actions and status of the interactive application collected during the current execution of the interactive application.

6 . The method according to claim 5 wherein the training of the parameters of the neural network takes into account a time difference between an occurrence of the first action and the obtaining of the first frame.

7 . The method according to claim 6 wherein, when the parameters of the neural network are trained offline, a plurality of sets of parameters are trained, each set of parameters being trained for a different value of offline time difference and wherein, during a current execution of the interactive application, the method comprises selecting the set of parameters of the plurality corresponding to the offline time difference the closest to an information representative of an actual time difference.

8 . The method according to claim 1 wherein the training of the parameters of the neural network uses a loss function estimating a difference between the second frame corresponding to the second action predicted by the neural network and a real frame generated by the interactive application corresponding to the same second action and wherein only a subpart of the second frame is displayed, only the displayed subpart being considered by the loss function.

9 . The method according to claim 1 wherein the interactive application is a network-based interactive application wherein the interactive application is managed by a remote equipment communicating with a local equipment via a network, the method being executed by the local equipment wherein:

the first action is performed by the user at a first time and registered by the local equipment and an information representative of the first action is transmitted to the remote equipment; and

the first frame and/or the metadata are obtained by decoding a portion of a video stream received from the remote equipment.

10 - 13 . (canceled)

14 . The method according to claim 9 wherein:

the first frame corresponds to a second action of the user at a second time following the first time predicted by the remote equipment from the information representative of the first action and information representative of a status of the interactive application at the first time;

and the method further comprises:

storing a reconstructed version of the first frame in a frame buffer used for temporal prediction of next frames;

receiving from the remote equipment a real frame corresponding to the second time after transmission to the remote equipment of data representative of an action performed by the user at the second time; and,

decoding the real frame and replacing the reconstructed version of the predicted frame by a reconstructed version of the real frame in the frame buffer.

15 . (canceled)

16 . A device for reducing a latency in an interactive application comprising electronic circuitry adapted for:

obtaining a first frame, the first frame being representative of a first action performed by a user in the interactive application;

obtaining information representative of a second action performed by the user in the interactive application, the second action following the first action; and,

predicting a second frame corresponding to the second action from data comprising at least the first frame and the information representative of a second action using a neural network.

17 . The device according to claim 16 wherein the electronic circuitry is further adapted for controlling a display of the second frame.

18 . The device according to claim 16 wherein the electronic circuitry is further adapted for obtaining metadata along with the first frame, the metadata being at least representative of a status of the interactive application at a time corresponding to the first action and/or of the first action, the second frame being further predicted from the metadata using the neural network.

19 . The device according to claim 18 wherein the metadata representative of a status of the interactive application comprise information representative of the user and/or information representative of dynamic objects and/or of other users in the interactive application.

20 . The device according to claim 16 wherein the neural network use parameters:

trained offline using data representative of frames, user actions and status of the interactive application collected during an offline execution of the interactive application; or,

trained on the fly using data representative of frames, user actions and status of the interactive application collected during a current execution of the interactive application; or,

initialized at a start of an execution of the interactive application using parameters trained offline using data representative of frames, user actions and status of the interactive application collected during an offline execution of the interactive application and then trained on the fly using data representative of frames, user actions and status of the interactive application collected during the current execution of the interactive application.

21 . The device according to claim 20 wherein the training of the parameters of the neural network takes into account a time difference between an occurrence of the first action and the obtaining of the first frame.

22 . The device according to claim 21 wherein, when the parameters of the neural network are trained offline, a plurality of sets of parameters are trained, each set of parameters being trained for a different value of offline time difference and wherein, during a current execution of the interactive application, the electronic circuitry is further adapted for selecting the set of parameters of the plurality corresponding to the offline time difference the closest to an information representative of an actual time difference.

23 . The device according to claim 16 wherein the training of the parameters of the neural network uses a loss function estimating a difference between the second frame corresponding to the second action predicted by the neural network and a real frame generated by the interactive application corresponding to the same second action and wherein only a subpart of the second frame is displayed, only the displayed subpart being considered by the loss function.

24 . The device according to claim 16 wherein the interactive application is a network-based interactive application wherein the interactive application is managed by a remote equipment communicating with the device via a network, the electronic circuitry being further adapted to:

register the first action, the first action being performed by a user at a first time;

transmit information representative of the first action to the remote equipment; and

obtaining the first frame and/or the metadata by decoding a portion of a video stream received from the remote equipment.

25 - 28 . (canceled)

29 . The device according to claim 24 wherein:

the first frame corresponds to a second action of the user at a second time following the first time predicted by the remote equipment from the information representative of the first action and information representative of a status of the interactive application at the first time; and the electronic circuitry is further adapted for:

storing a reconstructed version of the first frame in a frame buffer used for temporal prediction of next frames;

receiving from the remote equipment a real frame corresponding to the second time after transmission to the remote equipment of data representative of an action performed by the user at the second time; and,

decoding the real frame and replacing the reconstructed version of the predicted frame by a reconstructed version of the real frame in the frame buffer.

30 - 74 . (canceled)