US 11,956,447 B2
Using rate distortion cost as a loss function for deep learning
Claudionor Coelho, Redwood City, CA (US); Aki Kuusela, Palo Alto, CA (US); Joseph Young, Mountain View, CA (US); Shan Li, Fremont, CA (US); and Dake He, Sunnyvale, CA (US)
Assigned to GOOGLE LLC, Mountain View, CA (US)
Appl. No. 17/601,639
Filed by Google LLC, Mountain View, CA (US)
PCT Filed Mar. 21, 2019, PCT No. PCT/US2019/023339
§ 371(c)(1), (2) Date Oct. 5, 2021,
PCT Pub. No. WO2020/190297, PCT Pub. Date Sep. 24, 2020.
Prior Publication US 2022/0201316 A1, Jun. 23, 2022
Int. Cl. H04N 19/147 (2014.01); G06T 9/00 (2006.01); H04N 19/176 (2014.01); H04N 19/96 (2014.01)
CPC H04N 19/147 (2014.11) [G06T 9/002 (2013.01); H04N 19/176 (2014.11); H04N 19/96 (2014.11)] 20 Claims
OG exemplary drawing
 
1. An apparatus, comprising:
a processor that is configured to:
receive training data comprising:
a plurality of training blocks; and
for a training block of the plurality of training blocks:
a plurality of partition decisions used by an encoder for encoding the training block; and
for each partition decision of the plurality of partition decisions,
a rate-distortion value resulting from encoding the training block using the partition decision; and
train a machine-learning model to output a partition decision for encoding an image block by:
inputting the training data into a neural network using a loss function comprising a combination of:
a partition loss function that is based upon a relationship between the partition decisions and respective predicted partitions; and
a rate-distortion cost loss function that is based upon a relationship between the rate-distortion values and respective predicted rate-distortion values.