公开(公告)号：US10924629B1

公开(公告)日：2021-02-16

申请号：US16712330

申请日：2019-12-12

Applicant: Amazon Technologies, Inc.

Inventor： Christian Garcia Siagian ,  Christian Ciabattoni ,  Yang Yu ,  Yik Pui Suen ,  Ryan Barlow Dall ,  Ritesh Pase ,  Ramakanth Mudumba

IPC: H04N1/32 ,  G06K9/00 ,  G06N3/02 ,  H04N1/60

Abstract: Techniques for automated content validation are provided. In some examples, a media file and a metadata file associated with a title of the media file may be received. One or more scene paragraphs may be identified based at least in part on information in the metadata file. Scenes may be identified at least in part by using a transformer model implemented in a neural network. One or more scene files may be generated from the media file. One or more characters in a scene file of the one or more scene files may be identified. A match score may be determined, based at least in part on an association of the scene file to a scene paragraph of the plurality of scene paragraphs. A validity criterion may be determined for the title associated with the media file based at least in part on the match score.

公开(公告)号：US10904476B1

公开(公告)日：2021-01-26

申请号：US16712294

申请日：2019-12-12

Applicant: Amazon Technologies, Inc.

Inventor： Christian Garcia Siagian ,  Charles Effinger ,  David Niu ,  Yang Yu ,  Narayan Sundaram ,  Arjun Cholkar ,  Ramakanth Mudumba

IPC: H04N7/01 ,  G06K9/00 ,  G06T3/40

Abstract: Techniques for automated up-sampling of media files are provided. In some examples, a title associated with a media file, a metadata file associated with the title, and the media file may be received. The media file may be partitioned into one or more scene files, each scene file including a plurality of frame images in a sequence. One or more up-sampled scene files may be generated, each corresponding to a scene file of the one or more scene files. An up-sampled media file may be generated by combining at least a subset of the one or more up-sampled scene files. Generating one or more up-sampled scene files may include identifying one or more characters in a frame image of the plurality of frame images, based at least in part on implementation of a facial recognition algorithm including deep learning features in a neural network.