# Copyright (c) 2021-2025, InterDigital Communications, Inc
# All rights reserved.
# Redistribution and use in source and binary forms, with or without
# modification, are permitted (subject to the limitations in the disclaimer
# below) provided that the following conditions are met:
# * Redistributions of source code must retain the above copyright notice,
# this list of conditions and the following disclaimer.
# * Redistributions in binary form must reproduce the above copyright notice,
# this list of conditions and the following disclaimer in the documentation
# and/or other materials provided with the distribution.
# * Neither the name of InterDigital Communications, Inc nor the names of its
# contributors may be used to endorse or promote products derived from this
# software without specific prior written permission.
# NO EXPRESS OR IMPLIED LICENSES TO ANY PARTY'S PATENT RIGHTS ARE GRANTED BY
# THIS LICENSE. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND
# CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT
# NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
# PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR
# CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
# OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
# WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
# OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
# ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
from typing import Any, Dict, List, Mapping, Optional, Tuple
import torch.nn as nn
from torch import Tensor
from compressai.registry import register_module
from .base import LatentCodec
__all__ = [
"HyperpriorLatentCodec",
]
[docs]@register_module("HyperpriorLatentCodec")
class HyperpriorLatentCodec(LatentCodec):
"""Hyperprior codec constructed from latent codec for ``y`` that
compresses ``y`` using ``params`` derived from ``z_hat``.
Hyperprior entropy modeling introduced in
`"Variational Image Compression with a Scale Hyperprior"
<https://arxiv.org/abs/1802.01436>`_,
by J. Balle, D. Minnen, S. Singh, S.J. Hwang, and N. Johnston,
International Conference on Learning Representations (ICLR), 2018.
.. code-block:: none
┌───┐ z ┌──────┐ z_hat ┌───┐
┌─►─┤h_a├──►──┤ lc_z ├───►───┤h_s├─►─┐
│ └───┘ └──────┘ └───┘ │
│ ▼ params
│ │
│ ┌──┴───┐
y ──┴───────────────────►─────────────┤ lc_y ├───►── y_hat
└──────┘
The original hyperprior is the combination of an
entropy bottleneck for ``z`` and a gaussian conditional for ``y``:
.. code-block:: none
┌───┐ z ┌───┐ z_hat z_hat ┌───┐
┌─►──┤h_a├──►──┤ Q ├───►───····───►───┤h_s├──►─┐
│ └───┘ └───┘ EB └───┘ │
│ │
│ ┌──────────────◄────────────┘
│ │ params
│ ┌──┴──┐
│ │ EP │
│ └──┬──┘
│ │
│ ┌───┐ y_hat ▼
y ──┴─►─┤ Q ├────►────····────►── y_hat
└───┘ GC
Common configurations of latent codecs include:
- entropy bottleneck ``z`` and gaussian conditional ``y``
- entropy bottleneck ``z`` and autoregressive ``y``
"""
def __init__(
self,
latent_codec: Mapping[str, LatentCodec],
h_a: Optional[nn.Module] = None,
h_s: Optional[nn.Module] = None,
**kwargs,
):
super().__init__()
self.h_a = h_a or nn.Identity()
self.h_s = h_s or nn.Identity()
latent_codec = {
"y": latent_codec["y"],
"z": latent_codec.get("z") or latent_codec["hyper"],
}
self.y = latent_codec["y"]
self.z = latent_codec["z"]
self.latent_codec = latent_codec
def forward(self, y: Tensor) -> Dict[str, Any]:
z = self.h_a(y)
z_out = self.latent_codec["z"](z)
z_hat = z_out["y_hat"]
params = self.h_s(z_hat)
y_out = self.latent_codec["y"](y, params)
return {
"likelihoods": {
"y": y_out["likelihoods"]["y"],
"z": z_out["likelihoods"]["y"],
},
"y_hat": y_out["y_hat"],
}
def compress(self, y: Tensor) -> Dict[str, Any]:
z = self.h_a(y)
z_out = self.latent_codec["z"].compress(z)
z_hat = z_out["y_hat"]
params = self.h_s(z_hat)
y_out = self.latent_codec["y"].compress(y, params)
[z_strings] = z_out["strings"]
return {
"strings": [*y_out["strings"], z_strings],
"shape": {"y": y_out["shape"], "z": z_out["shape"]},
"y_hat": y_out["y_hat"],
}
def decompress(
self, strings: List[List[bytes]], shape: Dict[str, Tuple[int, ...]], **kwargs
) -> Dict[str, Any]:
*y_strings_, z_strings = strings
assert all(len(y_strings) == len(z_strings) for y_strings in y_strings_)
z_out = self.latent_codec["z"].decompress([z_strings], shape["z"])
z_hat = z_out["y_hat"]
params = self.h_s(z_hat)
y_out = self.latent_codec["y"].decompress(y_strings_, shape["y"], params)
return {"y_hat": y_out["y_hat"]}
