HDR10+ SEI payloadSize being injected twice.

Last version I was using for HDR10+ metadata encoding was from 2017-07-06 commit and the payloadSize was showing up correct in the payload stream.

Current version inserts the payloadSize into the stream TWICE and it actually has a +1 value for the first injected payloadSize value. (this +1 is the amount of bytes in the payload size value)

I think the issue comes from commit 2f6bb6f to fix Issue #353

From what I found, m_payloadSize is being used to write the payload size bytes at sei.cpp

    if (hrdTypes || m_payloadType == USER_DATA_UNREGISTERED || m_payloadType == USER_DATA_REGISTERED_ITU_T_T35)
    {
        if (hrdTypes)
        {
            X265_CHECK(0 == (count.getNumberOfWrittenBits() & 7), "payload unaligned\n");
            payloadSize = count.getNumberOfWrittenBits() >> 3;
        }
        else if (m_payloadType == USER_DATA_UNREGISTERED)
            payloadSize = m_payloadSize + 16;
        else
            payloadSize = m_payloadSize;

        for (; payloadSize >= 0xff; payloadSize -= 0xff)
            WRITE_CODE(0xff, 8, "payload_size");
        WRITE_CODE(payloadSize, 8, "payload_size");
    }
    else
        WRITE_CODE(m_payloadSize, 8, "payload_size");

but the actual m_payload from sei.h ALSO includes the payloadSize bytes, so they are writen twice.

    // daniel.vt@samsung.com :: for the Creative Intent Meta Data Encoding ( seongnam.oh@samsung.com )
    void writeSEI(const SPS&)
    {
        if (!m_payload)
            return;

        uint32_t i = 0;
        for (; i < m_payloadSize; ++i)
            WRITE_CODE(m_payload[i], 8, "creative_intent_metadata");
    }

The original idea was the second approach, but since being adapted to use m_payloadSize (the more libx265 like approach) there seems to be a conflict in the payloadSize injection logic.

It might be needed to skip the injection of payloadSize bytes from the m_payload in sei.h, and also don't consider that byte(s) for the value of m_payloadSize.

Comments (5)