US 12,488,579 B2
Aligning entities using neural networks
Antonia Phoebe Nina Creswell, London (GB)
Assigned to DeepMind Technologies Limited, London (GB)
Appl. No. 18/016,124
Filed by DeepMind Technologies Limited, London (GB)
PCT Filed Jul. 16, 2021, PCT No. PCT/EP2021/070023
§ 371(c)(1), (2) Date Jan. 13, 2023,
PCT Pub. No. WO2022/013441, PCT Pub. Date Jan. 20, 2022.
Claims priority of provisional application 63/052,878, filed on Jul. 16, 2020.
Prior Publication US 2023/0260271 A1, Aug. 17, 2023
Int. Cl. G06V 10/00 (2022.01); G06V 10/82 (2022.01); G06V 20/58 (2022.01)
CPC G06V 10/82 (2022.01) [G06V 20/58 (2022.01); G06V 2201/07 (2022.01)] 24 Claims
OG exemplary drawing
 
1. A method comprising:
obtaining respective current feature representations for each of a set of current entities that have been detected in an environment at a current time point;
obtaining respective historical feature representations for each of a set of historical entities that have been detected in the environment at one or more earlier time points preceding the current time point; and
processing an alignment input comprising (i) the respective historical feature representations for the set of historical entities and (ii) the current feature representations for the set of current entities using an alignment neural network to generate an alignment output that defines, for each current entity of one or more of the current entities, a corresponding historical entity that is the same as the current entity, wherein the alignment neural network includes a dynamics neural network and a permutation neural network and wherein processing the alignment input comprises:
processing a dynamics input comprising at least the historical feature representations using the dynamics neural network to generate a respective predicted feature representation for each of the set of historical entities that is predicted to characterize the historical entity at the current time point; and
processing (i) the respective predicted feature representations for the set of historical entities and (ii) the respective current feature representations for the set of current entities using the permutation neural network to generate the alignment output that defines, for each of one or more of the current entities, a corresponding historical entity that is the same as the current entity.