Content adaptive transform precision for video coding
Systems, methods, and instrumentalities are disclosed for obtaining coded video data comprising quantized transform coefficients for a plurality of blocks, obtaining a first precision factor associated with a first block for performing at least one decoding function on the first block, obtaining a second precision factor associated with a second block for performing the at least one decoding function on the second block, and performing the at least one decoding function on the quantized transform coefficients for the first block using the first precision factor and on the quantized transform coefficients for the second block using the second precision factor.
1 . An apparatus comprising a processor configured to perform:
obtaining coded video data comprising quantized transform coefficients for a plurality of blocks;
obtaining a first content-dependent precision factor associated with a first block for performing at least one decoding function on the first block, wherein the first content-dependent precision factor is obtained based on a first magnitude of the transform coefficients associated with a first source content associated with the first block independent of a first bit-depth associated with the first block, and wherein the first content-dependent precision factor is a first minimum shift associated with reducing the first magnitude of the transform coefficients to an inverse transform processing range;
obtaining a second content-dependent precision factor associated with a second block for performing the at least one decoding function on the second block, wherein the second content-dependent precision factor is obtained based on a second magnitude of the transform coefficients associated with a second source content associated with the second block independent of a second bit-depth associated with the second block, and wherein the second content-dependent precision factor is a second minimum shift associated with reducing the second magnitude of the transform coefficients to the inverse transform processing range; and
performing the at least one decoding function on the quantized transform coefficients for the first block using the first content-dependent precision factor and on the quantized transform coefficients for the second block using the second content-dependent precision factor.
2 . The apparatus of claim 1 , wherein the first content-dependent precision factor associated with the first block and the second content-dependent precision factor associated with the second block are further obtained from the coded video data.
3 . The apparatus of claim 1 , wherein the first content-dependent precision factor and the second content-dependent precision factor are different.
4 . The apparatus of claim 1 , wherein the at least one decoding function comprises dequantization or inverse transformation, the first content-dependent precision factor comprises a first dequantization shift, the second content-dependent precision factor comprises a second dequantization shift, and the first content-dependent precision factor and the second content-dependent precision factor comprise a mid-transform shift.
5 . The apparatus of claim 4 , wherein obtaining the first content-dependent precision factor associated with the first block comprises determining the mid-transform shift associated with the first block based on the bounds on an absolute magnitude of an output of a first inverse transform, and the at least one decoding function comprises a second inverse transform.
6 . The apparatus of claim 1 , wherein the first content-dependent precision factor comprises a dequantization shift and a mid-transform shift, and wherein the mid-transform shift associated with the first block is determined based on the dequantization shift associated with the first block.
7 . A method comprising:
obtaining coded video data comprising quantized transform coefficients for a plurality of blocks;
obtaining a first content-dependent precision factor associated with a first block for performing at least one decoding function on the first block, wherein the first content-dependent precision factor is obtained based on a first magnitude of the transform coefficients associated with a first source content associated with the first block independent of a first bit-depth associated with the first block, and wherein the first content-dependent precision factor is a first minimum shift associated with reducing the first magnitude of the transform coefficients to an inverse transform processing range;
obtaining a second content-dependent precision factor associated with a second block for performing the at least one decoding function on the second block, wherein the second content-dependent precision factor is obtained based on a second magnitude of the transform coefficients associated with a second source content associated with the second block independent of a second bit-depth associated with the second block, and wherein the second content-dependent precision factor is a second minimum shift associated with reducing the second magnitude of the transform coefficients to the inverse transform processing range; and
performing the at least one decoding function on the quantized transform coefficients for the first block using the first content-dependent precision factor and on the quantized transform coefficients for the second block using the second content-dependent precision factor.
8 . The method of claim 7 , wherein the first content-dependent precision factor associated with the first block and the second content-dependent precision factor associated with the second block are further obtained from the coded video data.
9 . The method of claim 7 , wherein the first content-dependent precision factor and the second content-dependent precision factor are different.
10 . The method of claim 7 , wherein the at least one decoding function comprises dequantization or inverse transformation, the first content-dependent precision factor comprises a first dequantization shift, the second content-dependent precision factor comprises a second dequantization shift, and the first content-dependent precision factor and the second content-dependent precision factor comprise a mid-transform shift.
11 . The method of claim 10 , wherein obtaining the first content-dependent precision factor associated with the first block comprises determining the mid-transform shift associated the first block based on the bounds on an absolute magnitude of an output of a first inverse transform, and the at least one decoding function comprises a second inverse transform.
12 . The method of claim 7 , wherein the first content-dependent precision factor comprises a dequantization shift and a mid-transform shift, and wherein the mid-transform shift associated with the first block is determined based on the dequantization shift associated with the first block.
13 . An apparatus comprising a processor configured to perform:
obtaining prediction residual data for a plurality of blocks of video data;
analyzing the prediction residual data to determine a magnitude input to represent a block;
determining a first content-dependent precision factor associated with a first block for performing at least one encoding function on the first block, wherein the first content-dependent precision factor is determined based on a first magnitude input associated with a first source content associated with the first block independent of a first bit-depth associated with the first block, and wherein the first content-dependent precision factor is a first minimum shift associated with reducing the first magnitude of the transform coefficients to an inverse transform processing range;
determining a second content-dependent precision factor associated with a second block for performing the at least one encoding function on the second block, wherein the second content-dependent precision factor is determined based on a second magnitude input associated a second source content associated with the second block independent of a second bit-depth associated with the second block, and wherein the second content-dependent precision factor is a second minimum shift associated with reducing the second magnitude of the transform coefficients to an inverse transform processing range; and
performing the at least one encoding function on the input for the first block using the first content-dependent precision factor and on the input for the second block using the second content-dependent precision factor.
14 . The apparatus of claim 13 , wherein the at least one encoding function comprises at least one of a horizontal transform or a quantization.
15 . The apparatus of claim 13 , wherein the processor is further configured to further perform:
include an indication of the first content-dependent precision factor for the first block and an indication of the second content-dependent precision factor for the second block in a bitstream representing the video data.
16 . The apparatus of claim 15 , wherein the indication of the first content-dependent precision factor comprises a number of fraction bits for performing at least one decoding function.
17 . The apparatus of claim 13 , wherein analyzing the prediction residual data to determine a largest magnitude input to represent a block is performed before vertical transformation.
18 . The apparatus of claim 13 , wherein the first content-dependent precision factor and the second content-dependent precision factor are different.