I’m decoding each RTP packet to see if it contains IDR NALU and when it does I pass NALUs (PPS, SPS, I-frame in the case) to decoding pipeline where I run NVIDIA Video Codec SDK

RTSP stream is h264, SVC off. So format selection for the decoder is out of question I believe

NvDecoder dec(cuContext, true, cudaVideoCodec_H264);
nFrameReturned = dec.Decode(nalu.data, nalu.len);

Decode function:

int NvDecoder::Decode(const uint8_t *pData, int nSize, int nFlags, int64_t nTimestamp)
{
    m_nDecodedFrame = 0;
    m_nDecodedFrameReturned = 0;
    CUVIDSOURCEDATAPACKET packet = { 0 };
    packet.payload = pData;
    packet.payload_size = nSize;
    packet.flags = nFlags | CUVID_PKT_TIMESTAMP;
    packet.timestamp = nTimestamp;
    if (!pData || nSize == 0) {
        packet.flags |= CUVID_PKT_ENDOFSTREAM;
    }
    NVDEC_API_CALL(cuvidParseVideoData(m_hParser, &packet));

    return m_nDecodedFrame;
}

However nFrameReturned is always zero, thus no frames are decoded

I suppose NALUs should be formatted in some way before being passed to the decoder?

I’ve tried flattening NALUs into 1d array, sending I-frame NALU only – the result is the same, no frames are decoded