Granted Patent B2

US 12,586,353 · App. 18/289,171 · Granted Mar 24, 2026

Inventors: Rishabh Kabra (London, GB); Daniel Zoran (London, GB); Goker Erdogan (London, GB); Antonia Phoebe Nina Creswell (London, GB); Loic Matthey-de-l'Endroit (London, GB); Matthew Botvinick (Philadelphia, PA); Alexander Lerchner (London, GB); Christopher Paul Burgess (Purley, GB)

Assignee: GDM Holding LLC

G06V10/771G06T9/00G06V10/44G06V10/82

Abstract

Claims (66)

1 . A computer-implemented method of training a video object representation neural network system, the video object representation neural network system including a self-attention neural network comprising one or more self-attention layers, each configured to apply a self-attention mechanism over a layer input, and having a plurality of self-attention neural network parameters, the method comprising:

obtaining a video sequence comprising a sequence of T image frames, one at each of a succession of times;

processing each of the image frames to generate a set of S spatial encodings for each of the T image frames thereby to obtain a group of S×T spatial encodings for the video sequence, wherein each spatial encoding of an image frame comprises an encoding of a respective region of the image frame;

processing the group of S×T spatial encodings using the self-attention neural network by applying the self-attention mechanism over a layer input comprising the group of S×T spatial encodings to generate a group of transformed spatial encodings each corresponding to an image frame time and to an image frame region, wherein applying the self-attention mechanism over the layer input includes applying the self-attention mechanism over the spatial encodings of different image frames;

determining, for each of a set of object latent variables, one or more values parameterizing an object latent distribution of the object latent variable, by aggregating the group of transformed spatial encodings over times of the image frames;

determining a value for each of the set of object latent variables by sampling from the object latent distribution for the object latent variable;

processing the determined values of the object latent variables to generate a decoded sequence of image frames; and

training the system by adjusting at least values of the self-attention neural network parameters to optimize an objective function dependent on i) a measure of a difference between the sequence of T image frames and the decoded sequence of image frames and ii) a difference between each of the object latent distributions and a prior object latent distribution.

2 . The method of claim 1 , further comprising:

determining, for each of a set of frame latent variables, one or more values parameterizing a frame latent distribution of the frame latent variable, by aggregating the group of transformed spatial encodings over regions of the image frames;

determining a value for each of the set of frame latent variables by sampling from the frame latent distribution for the frame latent variable;

wherein generating a decoded sequence of image frames further comprises processing the determined values of the frame latent variables to generate the decoded sequence of image frames.

3 . The method of claim 2 , wherein training the system comprises adjusting at least values of the self-attention neural network parameters to optimize the objective function dependent on a difference between each of the frame latent distributions and a prior frame latent distribution.

4 . The method of claim 1 , wherein each spatial encoding comprises a feature map of the respective region of the image frame, wherein processing each of the image frames to generate the set of S spatial encodings for each of the T image frames comprises processing each of the image frames using a feature extraction neural network having a plurality of feature extraction neural network parameters to generate the feature map for each region of the image frame; and wherein training the system includes adjusting values of the feature extraction neural network parameters.

5 . The method of claim 2 , further comprising using the trained video object representation neural network system to obtain, from the frame latent variables or from the values parameterizing the frame latent distributions, a position or orientation of a viewpoint for the video sequence.

6 . The method of claim 1 , comprising processing the determined values of the object latent variables using an image frame decoder neural network having a plurality of image frame decoder neural network parameters to generate the decoded sequence of image frames; and

wherein training the system further comprises adjusting values of the image frame decoder neural network parameters.

7 . The method of claim 6 wherein the decoded sequence of image frames comprises an image frame for each of a succession of decoded image frame time steps, and wherein using an image frame decoder neural network to generate the decoded sequence of image frames comprises, for each pixel of each decoded image frame and for each decoded image frame time step:

processing, using the image frame decoder neural network, the determined values of the object latent variables, information specifying a location of the pixel, and information specifying the time step to determine, for the pixel and for the time step, parameters of a pixel distribution for each of the object latent variables;

combining the pixel distributions for each of the object latent variables to determine a combined pixel distribution; and

sampling from the combined pixel distribution to determine a value for the pixel and for the time step.

8 . The method of claim 7 further comprising, for each pixel of each decoded image frame and for each decoded image frame time step:

processing, using the image frame decoder neural network, the determined values of the object latent variables, information specifying a location of the pixel, and information specifying the time step to determine, for the pixel and for the time step, a mixing weight for each of the object latent variables; and wherein

combining the pixel distributions for each of the object latent variables to determine a combined pixel distribution comprises combining the pixel distribution for each object latent variable weighted by the respective mixing weight.

9 . The method of claim 6 , further comprising:

obtaining a conditioning input defining one or more objects or object properties for one or more objects in the video sequence, or defining a viewpoint position or orientation; and

providing the conditioning input to the self-attention neural network and to the image frame decoder neural network.

10 . The method of claim 1 , wherein processing the group of S×T spatial encodings using the attention neural network comprises processing the layer input comprising the group of S×T spatial encodings using one of the one or more self-attention layers to generate a query and a key-value pair for each of the spatial encodings, and using the self-attention mechanism to apply the queries to the key-value pairs to determine the group of transformed spatial encodings; and wherein the self-attention neural network parameters comprise parameters of learned transformations applied to the layer input to generate the queries and key-value pairs.

11 . The method of claim 1 , wherein processing the group of S×T spatial encodings using the attention neural network comprises:

providing the group of S×T spatial encodings as the layer input to a first self-attention layer of the attention neural network to generate a first self-attention layer output; and

providing first self-attention layer output to a second self-attention layer of the attention neural network to generate the group of transformed spatial encodings; including

reducing the number of spatial encodings from S×T spatial encodings to K×T spatial encodings, where K is the number object latent variables and K is less than S.

12 . The method of claim 11 wherein the reducing comprises applying a pooling operation over the first self-attention layer output.

13 . The method of claim 1 , wherein training the system comprises backpropagating gradients of the objective function through an image frame decoder neural network generating the decoded sequence of image frames, the self-attention neural network, and a feature extraction neural network generating the set of spatial encodings.

14 . The method of claim 1 , further comprising using the trained video object representation neural network system to obtain one or more properties of one or more objects depicted in the video sequence from the object latent variables or from the values parameterizing the object latent distributions.

15 . The method of claim 1 , further comprising using the trained video object representation neural network system to generate a new video sequence comprising a decoded sequence of new image frames, by:

determining a value for each of the set of object latent variables by sampling from a prior object latent distribution for the object latent variable;

processing the determined values of the object latent variables to generate the decoded sequence of new image frames.

16 . The method of claim 15 , further comprising:

determining a value for each of the set of frame latent variables by sampling from a prior frame latent distribution for the frame latent variable; and

processing the determined values of the object latent variables and the determined values of the frame latent variables to generate the decoded sequence of new image frames.

17 . The method of claim 1 , further comprising using the trained video object representation neural network system to generate a modified video sequence comprising a decoded sequence of modified image frames, by:

obtaining an input video sequence comprising a sequence of input image frames;

processing each of the input image frames to obtain a group of spatial encodings for the input video sequence;

processing the group of spatial encodings for the input video sequence using the attention neural network to generate a group of transformed spatial encodings;

determining values for the set of object latent variables by sampling from object latent distributions determined from the group of transformed spatial encodings;

modifying the determined values for the set of object latent variables, to obtain a modified set of latent variables; and

processing the modified set of latent variables to generate the decoded sequence of modified image frames.

18 . The method of claim 17 , further comprising determining values for the set of frame latent variables by sampling from frame latent distributions determined from the group of transformed spatial encodings;

modifying the determined values for the set of frame latent variables, to obtain the modified set of latent variables.

19 . One or more non-transitory computer-readable storage media storing instructions that when executed by one or more computers cause the one or more computers to perform operations for training a video object representation neural network system, the video object representation neural network system including a self-attention neural network comprising one or more self-attention layers, each configured to apply a self-attention mechanism over a layer input, and having a plurality of self-attention neural network parameters, the operations comprising:

obtaining a video sequence comprising a sequence of T image frames, one at each of a succession of times;

processing each of the image frames to generate a set of S spatial encodings for each of the T image frames thereby to obtain a group of S×T spatial encodings for the video sequence, wherein each spatial encoding of an image frame comprises an encoding of a respective region of the image frame;

processing the group of S×T spatial encodings using the self-attention neural network by applying the self-attention mechanism over a layer input comprising the group of S×T spatial encodings to generate a group of transformed spatial encodings each corresponding to an image frame time and to an image frame region, wherein applying the self-attention mechanism over the layer input includes applying the self-attention mechanism over the spatial encodings of different image frames;

determining, for each of a set of object latent variables, one or more values parameterizing an object latent distribution of the object latent variable, by aggregating the group of transformed spatial encodings over times of the image frames;

determining a value for each of the set of object latent variables by sampling from the object latent distribution for the object latent variable;

processing the determined values of the object latent variables to generate a decoded sequence of image frames; and

training the system by adjusting at least values of the self-attention neural network parameters to optimize an objective function dependent on i) a measure of a difference between the sequence of T image frames and the decoded sequence of image frames and ii) a difference between each of the object latent distributions and a prior object latent distribution.

20 . A system comprising one or more computers and one or more storage devices storing instructions that when executed by one or more computers cause the one or more computers to perform operations for training a video object representation neural network system, the video object representation neural network system including a self-attention neural network comprising one or more self-attention layers, each configured to apply a self-attention mechanism over a layer input, and having a plurality of self-attention neural network parameters, the operations comprising:

obtaining a video sequence comprising a sequence of T image frames, one at each of a succession of times;

processing each of the image frames to generate a set of S spatial encodings for each of the T image frames thereby to obtain a group of S×T spatial encodings for the video sequence, wherein each spatial encoding of an image frame comprises an encoding of a respective region of the image frame;

processing the group of S×T spatial encodings using the self-attention neural network by applying the self-attention mechanism over a layer input comprising the group of S×T spatial encodings to generate a group of transformed spatial encodings each corresponding to an image frame time and to an image frame region, wherein applying the self-attention mechanism over the layer input includes applying the self-attention mechanism over the spatial encodings of different image frames;

determining, for each of a set of object latent variables, one or more values parameterizing an object latent distribution of the object latent variable, by aggregating the group of transformed spatial encodings over times of the image frames;

determining a value for each of the set of object latent variables by sampling from the object latent distribution for the object latent variable;

processing the determined values of the object latent variables to generate a decoded sequence of image frames; and

training the system by adjusting at least values of the self-attention neural network parameters to optimize an objective function dependent on i) a measure of a difference between the sequence of T image frames and the decoded sequence of image frames and ii) a difference between each of the object latent distributions and a prior object latent distribution.

Assignments (2)

Continuity (2)

References Cited (78)