Back to Search
Document details
Non-CE2: Motion vector clipping in affine sub-block motion vector derivation
Abstract
In affine motion prediction, the motion field is derived for each sub-block based on the control point motion vectors. Even if the control point motion vectors are in the range specified by the motion field storage bit depth, the derived sub-block motion vectors may be outside of the range of values that can be represented within the motion field bit depth, resulting in arithmetic underflow and/or overflow. However, no motion vector clipping to the motion field storage bit depth (e.g., 16-bit in VTM) is considered in the VTM-3.0 software nor in the current VVC draft. To ensure the same behavior between different implementations and to avoid encoder crash in some cases, it is proposed to apply clipping to motion field storage bit depth to both sub-block motion vectors and control point motion vectors after the sub-block motion vector derivation in affine mode. Simulation results show that the proposed fix provides similar performance as VTM-3.0 in terms of BD-rate and timing.
JVET-M0145 Non-CE2: Motion vector clipping in affine sub-block motion vector derivation [P. Hanhart, Y. He (InterDigital)]
References:
Decisions
JVET-M0145 Non-CE2: Motion vector clipping in affine sub-block motion vector derivation [P. Hanhart, Y. He (InterDigital)]
Citation