公开(公告)号：US20230362367A1

公开(公告)日：2023-11-09

申请号：US18333067

申请日：2023-06-12

Applicant: SAMSUNG ELECTRONICS CO., LTD.

Inventor： Anubhav SINGH ,  Aviral AGRAWAL ,  Raj Narayana GADDE ,  H Keerthan BHAT ,  Yinji PIAO ,  Minwoo PARK ,  Kwangpyo CHOI

IPC: H04N19/119 ,  H04N19/117 ,  H04N19/80 ,  H04N19/42 ,  H04N19/176

CPC classification number: H04N19/119 ,  H04N19/117 ,  H04N19/80 ,  H04N19/42 ,  H04N19/176

Abstract: An example method for training AI models for in-loop filters includes generating a training dataset by passing a video through a codec pipeline, extracting one or more predefined block features from the training dataset, creating a plurality of clusters based on the extracted one or more predefined block features from the training dataset, dividing the plurality of clusters into a sub-plurality of clusters based on the extracted one or more predefined block features and an intra-cluster variation threshold, and supplying the sub-plurality of clusters separately into a plurality of AI models based on the extracted one or more predefined block features.

公开(公告)号：US20230281458A1

公开(公告)日：2023-09-07

申请号：US18174976

申请日：2023-02-27

Applicant: Samsung Electronics Co., Ltd.

Inventor： Aviral AGRAWAL ,  Raj Narayana GADDE ,  Anubhav SINGH ,  Yinji PIAO ,  Minwoo PARK ,  Kwangpyo CHOI

IPC: G06N3/091 ,  G06T9/00

CPC classification number: G06N3/091 ,  G06T9/002

Abstract: A method and an electronic device for low-complexity in-loop filter inference using feature-augmented training are provided. The method includes combining spatial and spectral domain features, using spectral domain features for global feature extraction and signalling to the spatial stream during training, using a detachable spectral domain stream for differential complexity during training versus inference, and combining a unique set of losses resulting from multi-stream and multi-feature approaches to obtain an optimal output.