id	summary	reporter	owner	description	type	status	priority	milestone	component	version	resolution	keywords	cc
687	P1034 - Wrap around in scaling list coding	bheng		"For scaling list coding, there seems to be a wraparound process missing for the nextCoef derivation.

Specifically, the nextCoef is an accumulation of delta values, which could be as small as -128 each time:

    nextCoef += scaling_list_delta_coef[ id ][ i ]

There is no wrapping process here, so after accumulating a few coefficients, nextCoef could be a very large negative number.

ScalingList[id][i] is just set equal to nextCoef.

    ScalingList[ id ][ i ] = nextCoef

The value of ScalingList[id][i] is added to the predicted value and modulo wrapped:

    ScalingMatrixRec[ id ][ x ][ y ] = ( ScalingMatrixPred[ x ][ y ] + ScalingList[ id ][ k ] + 256 ) % 256 )    (7-72)


The wrapping used here assumes that the (ScalingMatrixPred + ScalingList) sum is no smaller than -256.  However, as shown above, ScalingList could be a very large negative value. That would require a modulo of a negative number, which is undefined in the spec.

For example, the sequence of coefficients [144, 16, 144, 16, 144, 16, ... ] in scan order, requires all scaling_list_delta_coef = -128.  It's impossible to code this sequence without either creating a negative modulo situation above, or coding delta values outside the legal range [-128,127] (like VTM does).

Assuming it was not intended to make certain patterns of scaling lists impossible to encode, I believe the nextCoeff value should be wrapped to stay within [-128,127].

Also, there seems to be a left parenthesis missing in equation (7-72) above.



"	defect	closed	minor		spec	VVC D7 vE	fixed		ksuehring bbross XiangLi fbossen jvet@…