Skip to content

Fix DebugBitstream issue caused by PDP

Gleb Verba requested to merge gverba/ECM:FIX-DEBUG-BITSTREAM-PDP into master

Bug

The bug can be reproduced following the workflow for RaceHorses CTC RA QP27 GOP04:

  1. Encode:

    EncoderApp.exe -c encoder_randomaccess_ecm.cfg -c classD_randomaccess.cfg -c RaceHorses.cfg --SEIDecodedPictureHash=1 --PrintHexPSNR=1 --FramesToBeEncoded=33 --IntraPeriod=32 --FrameSkip=128 -q 27 -o rec.yuv --Verbosity=6 -b encoded.bin

  2. Read 2 frames and start encoding from POC=16:

    EncoderApp.exe -c encoder_randomaccess_ecm.cfg -c classD_randomaccess.cfg -c RaceHorses.cfg --SEIDecodedPictureHash=1 --PrintHexPSNR=1 --FramesToBeEncoded=33 --IntraPeriod=32 --FrameSkip=128 -q 27 -o rec.yuv --Verbosity=6 -b recoded.bin --DebugBitstream=encoded.bin --DebugPOC=16

Segmentation fault at POC 16

Safety

No encoding behavior change is observed for CTC

Result

No bitstream issues under workflow Encode - Rewrite - Decode for CTC (not affected by this commit)

Though exhaustive testing is not feasible, a large number of tests has been executed and no issue is observed under workflow Encode - Read N frames and Encode rest - Decode for CTC (after this commit)

In simpler words, all DebugBitstream issues seem to be solved at the moment.

Merge request reports