Activation function design in neural network-based filtering process for video coding
A method of coding video data, the method comprising: reconstructing a block of the video data; and applying a Convolutional Neural Network (CNN)-based filter to the reconstructed block, wherein the CNN-based filter uses a LeakyReLU activation function.
1. A method of encoding video data, the method comprising:
reconstructing a block of the video data;
determining a plurality of values of an Alpha parameter;
applying a Convolutional Neural Network (CNN)-based filter to the reconstructed block, wherein:
the CNN-based filter uses a Leaky Rectified Linear Unit (LeakyReLU) activation function,
the LeakyReLU activation function is defined as:
f
(
y
)
=
{
Alpha
*
y
,
y
<
0
y
,
y
≥
0
y is an output value of a convolutional layer of a CNN of the CNN-based filter,
the CNN includes a plurality of convolutional layers, and
applying the CNN-based filter to the reconstructed block comprises using different values of the Alpha parameter from the plurality of values of the Alpha parameter in two or more different convolutional layers of the plurality of convolutional layers; and
encoding, in a bitstream that comprises an encoded representation of the video data, one or more syntax elements that provide information for deriving the plurality of values of the Alpha parameter at a video decoder.
2. The method of claim 1 , wherein the one or more syntax elements directly specify the values of the Alpha parameter.
3. The method of claim 1 , wherein the one or more syntax elements specify indexes corresponding to the values of the Alpha parameter within a predefined set of indices.
4. A device for encoding video data, the device comprising:
a memory to store the video data; and
one or more processors implemented in circuitry, the one or more processors configured to:
reconstruct a block of the video data;
determine a plurality of values of an Alpha parameter;
apply a Convolutional Neural Network (CNN)-based filter to the reconstructed block, wherein:
the CNN-based filter uses a Leaky Rectified Linear Unit (LeakyReLU) activation function,
the LeakyReLU activation function is defined as:
f
(
y
)
=
{
Alpha
⋆
y
,
y
<
0
y
,
y
≥
0
y is an output value of a convolutional layer of a CNN of the CNN-based filter,
the CNN includes a plurality of convolutional layers, and
the one or more processors are configured to, as part of applying the CNN-based filter to the reconstructed block, use different values of the Alpha parameter from the plurality of values of the Alpha parameter in two or more different convolutional layers of the plurality of convolutional layers; and
encode, in a bitstream that comprises an encoded representation of the video data, one or more syntax elements that provide information for deriving the plurality of values of the Alpha parameter at a video decoder.
5. The device of claim 4 , wherein the device comprises one or more of a camera, a computer, a mobile device, a broadcast receiver device, or a set-top box.
6. The device of claim 4 , wherein the device comprises a video encoder.
7. A device for encoding video data, the device comprising:
means for reconstructing a block of the video data;
means for determining a plurality of values of an Alpha parameter;
means for applying a Convolutional Neural Network (CNN)-based filter to the reconstructed block, wherein:
the CNN-based filter uses a Leaky Rectified Linear Unit (LeakyReLU) activation function,
the LeakyReLU activation function is defined as:
f
(
y
)
=
{
Alpha
⋆
y
,
y
<
0
y
,
y
≥
0
y is an output value of a convolutional layer of a CNN of the CNN-based filter,
the CNN including a plurality of convolutional layers, and
applying the CNN-based filter to the reconstructed block comprises using different values of the Alpha parameter from the plurality of values of the Alpha parameter in two or more different convolutional layers of the plurality of convolutional layers; and
means for encoding, in a bitstream that comprises an encoded representation of the video data, one or more syntax elements that provide information for deriving the plurality of values of the Alpha parameter at a video decoder.
8. A non-transitory computer-readable storage medium having stored thereon instructions that, when executed, cause one or more processors to:
reconstruct a block of video data;
determine a plurality of values of an Alpha parameter;
apply a Convolutional Neural Network (CNN)-based filter to the reconstructed block, wherein:
the CNN-based filter uses a Leaky Rectified Linear Unit (LeakyReLU) activation function,
the LeakyReLU activation function is defined as:
f
(
y
)
=
{
Alpha
⋆
y
,
y
<
0
y
,
y
≥
0
y is an output value of a convolutional layer of a CNN of the CNN-based filter,
the CNN including a plurality of convolutional layers, and
applying the CNN-based filter to the reconstructed block comprises using different values of the Alpha parameter from the plurality of values of the Alpha parameter in two or more different convolutional layers of the plurality of convolutional layers; and
encode, in a bitstream that comprises an encoded representation of the video data, one or more syntax elements that provide information for deriving the plurality of values of the Alpha parameter at a video decoder.
9. A method of decoding video data, the method comprising:
reconstructing a block of the video data; and
determining a value of an Alpha parameter based on one or more syntax elements signaled in a bitstream that comprises an encoded representation of the video data; and
applying a Convolutional Neural Network (CNN)-based filter to the reconstructed block, wherein the CNN-based filter uses a Leaky Rectified Linear Unit (LeakyReLU) activation function, wherein the LeakyReLU activation function is defined as:
f
(
y
)
=
{
Alpha
⋆
y
,
y
<
0
y
,
y
≥
0
where y is an output value of a convolutional layer of a CNN of the CNN-based filter and the Alpha parameter is a fixed parameter.
10. The method of claim 9 , wherein the value of the Alpha parameter is signaled in the bitstream.
11. The method of claim 9 ,
wherein an index of the Alpha parameter within a predefined set is signaled in the bitstream, and
wherein determining the value of the Alpha parameter comprises determining the value of the Alpha parameter in the predefined set to which the index of the Alpha parameter corresponds.
12. A device for decoding video data, the device comprising:
a memory to store the video data; and
one or more processors implemented in circuitry, the one or more processors configured to:
reconstruct a block of the video data; and
determine a value of an Alpha parameter based on one or more syntax elements signaled in a bitstream that comprises an encoded representation of the video data; and
apply a Convolutional Neural Network (CNN)-based filter to the reconstructed block, wherein the CNN-based filter uses a Leaky Rectified Linear Unit (LeakyReLU) activation function, wherein the LeakyReLU activation function is defined as:
f
(
y
)
=
{
Alpha
⋆
y
,
y
<
0
y
,
y
≥
0
where y is an output value of a convolutional layer of a CNN of the CNN-based filter and the Alpha parameter is a fixed parameter.
13. The device of claim 12 , wherein the value of the Alpha parameter is signaled in the bitstream.
14. The device of claim 12 ,
wherein an index corresponding to the value of the Alpha parameter within a predefined set of indices is signaled in the bitstream, and
wherein determining the value of the Alpha parameter comprises determining the value of the Alpha parameter in the predefined set to which the index of the Alpha parameter corresponds.
15. A method of encoding video data, the method comprising:
reconstructing a block of the video data; and
determining a single value of an Alpha parameter;
applying a Convolutional Neural Network (CNN)-based filter to the reconstructed block, wherein the CNN-based filter uses a Leaky Rectified Linear Unit (LeakyReLU) activation function, wherein the LeakyReLU activation function is defined as:
f
(
y
)
=
{
Alpha
⋆
y
,
y
<
0
y
,
y
≥
0
where y is an output value of a convolutional layer of a CNN of the CNN-based filter and the Alpha parameter is a fixed parameter; and
encoding, in a bitstream that comprises an encoded representation of the video data, one or more syntax elements that provide information for deriving the value of the Alpha parameter at a video decoder.
16. The method of claim 15 , wherein the one or more syntax elements directly specify the value of the Alpha parameter.
17. The method of claim 15 , wherein the one or more syntax elements specify an index corresponding to the value of the Alpha parameter within a predefined set of indices.
18. A device for encoding video data, the device comprising:
a memory to store the video data; and
one or more processors implemented in circuitry, the one or more processors configured to:
reconstruct a block of the video data; and
determine a single value of an Alpha parameter;
apply a Convolutional Neural Network (CNN)-based filter to the reconstructed block, wherein the CNN-based filter uses a Leaky Rectified Linear Unit (LeakyReLU) activation function, wherein the LeakyReLU activation function is defined as:
f
(
y
)
=
{
Alpha
⋆
y
,
y
<
0
y
,
y
≥
0
where y is an output value of a convolutional layer of a CNN of the CNN-based filter and the Alpha parameter is a fixed parameter; and
encode, in a bitstream that comprises an encoded representation of the video data, one or more syntax elements that provide information for deriving the value of the Alpha parameter at a video decoder.