Method for learned image compression and related autoencoder

Patent metadata

Field	Value
Publication number	US-12587664-B2
Application number	US-202418758390-A
Country	US
Kind code	B2
Filing date	Jun 28, 2024
Priority date	Sep 11, 2023
Publication date	Mar 24, 2026
Grant date	Mar 24, 2026

Abstract

A method for learned image compression implemented in an autoencoder includes: a) extracting from an image a latent space by the learnable encoder; b) quantizing the latent space by a quantizer to obtain a quantized latent space; c) entropy coding the quantized latent space by an entropy encoder to obtain a bitstream, wherein an entropy model used to encode the latent space is represented by a probability distribution; d) entropy decoding the bitstream by an entropy decoder to obtain an entropy decoded bitstream; e) feeding the entropy decoded bitstream to the decoder; f) recover a reconstructed image by the decoder; g) training the autoencoder via standard gradient descent of the backpropagated error gradient by finding learnable parameters of the learnable encoder and of the decoder that minimize a rate distortion cost function, wherein the entropy encoder is based on a differentiable formulation of a soft frequency counter.

First claim

The invention claimed is:

1 . A method for learned image compression implemented in an autoencoder comprising a learnable encoder (f a ) and a decoder (f s ), said method comprising the steps of:
a) extracting from an image (x) a latent space (y) by means of said learnable encoder (f a ); b) quantizing said latent space (y) by means of a quantizer (U|Q) to obtain a quantized latent space (ŷ); c) entropy coding said quantized latent space (ŷ) by means of an entropy encoder to obtain a bitstream, wherein an entropy model used to encode said latent space (y) is represented by a probability distribution p ŷ ; d) entropy decoding said bitstream by means of an entropy decoder to obtain an entropy decoded bitstream; e) feeding said entropy decoded bitstream to said decoder (f s ); f) recover a reconstructed image ({circumflex over (x)}) by means of said decoder (f s ); g) training said autoencoder via standard gradient descent of the backpropagated error gradient by finding learnable parameters (θ f ,θ g ) of said learnable encoder (f a ) and of said decoder (f s ) that minimize a rate distortion cost function L, wherein said entropy encoder is based on a differentiable formulation of a soft frequency counter (SFC).

2 . The method according to claim 1 , wherein said latent space (y) comprises a number N c of latent space channels having a dimension N d , and wherein, given a j-th channel of said latent space (y) and a quantization level l i j of said j-th channel, the soft frequency counter (SFC) associates every value of said latent space y n j to a weight inversely proportional to the distance with l i j , where n varies within the same channel and ranges from 1 to N d .

3 . The method according to claim 2 , wherein said soft frequency counter (SFC) relies on a scalar function φ i j and wherein a first order entropy H {tilde over (p)} of a probability distribution {acute over (p)} j for every single channel of said latent space is:

H

p
~

=

-

1

N
c

⁢

∑

j
=
1

N
c

H

p
~

j

=

-

1

N
c

⁢

∑

j
=
1

N
c

∑

i
=
1

L

SFC
⁡
(

l
i
j

)

⁢

log
2

[

SFC
⁡
(

l
i
j

)

]

where

SFC
⁡
(

l
i
j

)

=

∑

n
=
1

N
d

ϕ
i
j

(

y
n
j

)

∑

m
=
1

L

∑

n
=
1

N
d

ϕ
m
j

(

y
n
j

)

and

ϕ
i
j

Assignees

• Sisvel Tech S R L
• Univ Degli Studi Di Torino
• Inst Mines Telecom

Inventors

• Presta Alberto
• Fiandrotti Attilio
• Tartaglione Enzo
• Grangetto Marco

Classifications

• G06N3/0455 Primary

  Auto-encoder networks; Encoder-decoder networks · CPC title

• H04N19/91

  Entropy coding, e.g. variable length coding [VLC] or arithmetic coding · CPC title

• H04N19/124

  Quantisation · CPC title

• H04N19/147

  according to rate distortion criteria (rate-distortion as a criterion for motion estimation H04N19/567) · CPC title

• H04N19/13

  Adaptive entropy coding, e.g. adaptive variable length coding [AVLC] or context adaptive binary arithmetic coding [CABAC] · CPC title