The H.264 platinum codecs feature support for "Adaptive Longterm Frame Insertion", see here:
http://processors.wiki.ti.com/index.php/H.264_DM36x_Ver_2.0_Codec#Adaptive_Longterm_Frame_Insertion
I'm attempting to use this feature in a DM368-based system but I'm not sure I'm setting it up correctly. With the feature turned off and a simulated packet loss of 0.1% the decoder behaves as expected. I get a decode error on the frame with the dropped packet and then every frame after that reports extended error code 0x4f "The current picture refers to an already decoded error frame" ...until the next I-frame comes along and stops the error being propagated. All as expected.
With the feature enabled the reporting from the decoder looks correct. I still get the initial error from the frame with the dropped packet, then a couple of 0x4f warnings and then they stop. The error propagation appears to have been halted by instructing the encoder to "UseLongTermFrame=1". However the picture looks awful. It looks to me like setting UseLongTermFrame=1 on the encoder immediately corrupts the image. The result is much worse than the original error left to propagate. I'm assuming I'm not setting LongTermFrames correctly, or there's a mismatch between the references used by encode and decode.
My settings are;
EnableLongTermFrame = 1;
LongTermRefreshInterval = 30;
UseLongTermFrame is set to 1 whenever the decoder signals an error
SetLongTermFrame = 0; // Do I have to set this explicitly if I have set the refresh interval?
I've also tried setting LongTermRefreshInterval = 0 because the user guide says that I frames are automatically counted as long term frames. However this makes no difference.
Hope someone can shed some light on the right way to use this feature.
Alex
Hi Alex,
It looks like decoder is not getting the correct reference. Are you allocating sufficient number of buffers?
You can also check with decoder application in the package app/client/test/src/TestappDecoder_ih264_arm926.c. We have tested your stream with this application.
You can also refer to the link for more about buffer mechanism
http://www.ti.com/lit/an/sprab88/sprab88.pdf
Hi Adithya,
Thanks for your help here. I've marked it as verified because it was our mistake with buffer management that was causing the corruption. The link you provided made it clear what we needed to do to fix.
However running in universal mode is bad news for us as it has a significant impact on performance. I thought the idea of closed loop was that it could decode anything the TI encoder produced That is clearly not the case if the decoder can't decode long term reference frames with closedloop=1. Please could we raise a bug report/feature request against this?
Many thanks,
We will look into this but please dont expect any near term fix. Closed loop decoder was meant to decode streams of ver 2.0 encoder. We added more features in later version of encoder but the closed loop decoder did not get upgraded.
regards
Yashwant
Hi Yashwant,
Thanks for the honest assessment. I see that the last release for the decoder was almost a year ago. Shouldn't a new release be due by now?
Also, is there anything I can do to speed up the universal decoder since I'm not using lots of other features it has? Perhaps by specifying a lower levelIdc in the encoder?
All the best,