Datapoints for h.265

Hi

I have been using h265 videos as much as i could to find the quirks when using it and also retain better quality for longer videos. Here is what i found so far.

I mainly use handbrake to reencode videos and here is what i found. When i use GPU encode then artifacting happenes about 80% of the time. If i use CPU encode artifacting happens 10% of the time (or less).

Last week i uploaded a 4k video and it started artifacting on the 2nd loop. The video ended with a black screen and started with a black screen with minimal text (this is a clue).
If i inserted a colored picture in between the the video loop. The artifacting went away. If i tried the same with a black image, the artifacting came back. (the artifacting was green, never seen any other color artifacting)

Yesterday that device stopped working. It had the signs of overheating, however when i look at the history. Temps were fine. But memory kept climbing in a weird way.

Could it be that the decode process gets a memory leak or overflow dependent on the what goes on in the image? It feels like a Malloc is off by 2 bytes or something when compressibility is higher er then expected (as is with a mostly 5-10sec video frames).

Interesting and strange. Would it be possible to capture what you see somehow, so I might get an idea what’s happening?

Which device is that? It might make sense to first install the latest OS, to ensure this isn’t some issue with an older OS version.

The memory usage is expected to look like this: Video playback results in a bit of memory fragmentation and it takes a moment before it stabilizes to what’s actually needed. 200MB usage is pretty low, so that’s all good.

Here is the data about the device. Its a pi4.
I can change the setup to induce the artifacting when there are less visitors around. All my devices are in use at the moment so i can not set up a test screen. It may take up to a week for me to find a suitable time.

Ok. It’s now on the latest OS, so you really shouldn’t see any issues. If you do, I’ll see if I can reproduce this on a local Pi4. Let me know.