Artifacts in a sample with delayed random access points
They can be seen in the frames right after said points.
I'm not entirely sure if the sample is spec compliant, for that matter. I made it today with aomenc git head, and noticed that while section 7.6.2 states that the Sequence Header should be in the same TU as the coded key frame, in here it's in the same TU as the recovery point (frame with show_existing_frame pointing to the delayed key frame). It shouldn't affect decoding, i think, but it caught my attention.