In the RateCtrlMetaData implementation there is a reference to the Motion Vector buffer that is allocated and assigned for the VCU. I've dumped the data from the buffer to analyze it, and attempt to cross reference it with the motion vectors that are extracted by FFMpeg from the encoded stream.
The data in the motion vector buffer from the VCU seems very sparse, and I can't seem to make any sense of it. The VCU manual references the motion vector buffer several times, but doesn't give a clear explanation or provide a memory map to describe the structure of the data in the buffer.
It would be incredibly useful to be able to grab the motion vector data directly from the VCU rather than decoding it. It would significantly reduce the CPU overhead required for certain image processing tasks as motion data would be available directly from the fabric.
Hello,
In the
RateCtrlMetaDataimplementation there is a reference to the Motion Vector buffer that is allocated and assigned for the VCU. I've dumped the data from the buffer to analyze it, and attempt to cross reference it with the motion vectors that are extracted by FFMpeg from the encoded stream.The data in the motion vector buffer from the VCU seems very sparse, and I can't seem to make any sense of it. The VCU manual references the motion vector buffer several times, but doesn't give a clear explanation or provide a memory map to describe the structure of the data in the buffer.
It would be incredibly useful to be able to grab the motion vector data directly from the VCU rather than decoding it. It would significantly reduce the CPU overhead required for certain image processing tasks as motion data would be available directly from the fabric.
Thanks in advance!