Systems and methods to adaptively select long term reference frames for streaming video
AssigneeAmazon Technologies, Inc.
InventorSrikkanth Jagannathan
Life of this patent
Estimating the term…
Abstract
Systems and methods to adaptively select long term reference (LTR) frames in order to maintain video quality for streaming video services may comprise an encoder of a video server, a decoder of a client device, and a network. The encoder may determine a frequency for marking frames as LTR frames based on various factors, and may select, encode, and transmit LTR frames based on the determined frequency. In addition, the encoder may determine a content or quality metric of encoded frames, and may select, encode, and transmit LTR frames based on the content or quality metric.
Claims17
A method, comprising:
- determining, by an encoder of a video server, a frequency for marking frames of a video stream as long term reference (LTR) frames based at least in part on a network latency between the encoder and a decoder of a client device and an available number of LTR slots associated with the decoder, the video stream comprising a first set of frames;
- selecting, by the encoder from the first set of frames, a second set of frames to mark as the LTR frames at the frequency, wherein selecting the second set of frames comprises:
- determining, by the encoder, a content difference between an encoded frame and an acknowledged LTR frame;
- determining, by the encoder, that the content difference is greater than a threshold value;
- and
- selecting, by the encoder, a next frame of the first set of frames to mark as a next LTR frame based on the content difference;
- encoding, by the encoder, the first set of frames including the second set of frames;
- and
- transmitting, by the encoder via a first communication device, the first set of frames including the second set of frames to the decoder via a network interface.
A method, comprising:
- determining, by an encoder of a video server, a frequency for marking frames of a video stream as long term reference (LTR) frames based at least in part on a network latency between the encoder and a decoder of a client device, the video stream comprising a first set of frames;
- selecting, by the encoder from the first set of frames, a second set of frames to mark as the LTR frames at the frequency, wherein selecting the second set of frames comprises:
- determining, by the encoder, a content difference between an encoded frame and an acknowledged LTR frame;
- determining, by the encoder, that the content difference is greater than a threshold value;
- and
- selecting, by the encoder, a next frame of the first set of frames to mark as a next LTR frame based on the content difference;
- and
- encoding and transmitting, by the encoder, the first set of frames including the second set of frames to the decoder via a network interface.
Dependencies are inferred from the claim text.