Note the RADL between -1 and -7. 7 extra frames were inserted at that point for random access. Sequential access with zero random access is correct all the way to the end.
The source is cut from a TV stream, so the very first IDR is actually close to the end of the program. If I truncate the file to remove IDR and following stream, the file is indexed correctly. So it's a few RADLs that were supposed to be positioned around the IDR that were moved to the incorrect places.
I'll do a quick POC re-organizing the RADLs to their correct positions to see if that solves the problem.
Thank you for the stream and your analysis. I am currently getting only 100 KB/s download speed from my host. So it will unfortunately be a few days before I can even look at it. Why do I pay for 100 MB/s internet connection?
Got the stream, thank you. I know what's going on but I'm in the middle of a kitchen tile project (hard to stop in midstream). Ever start a project and wish you hadn't?
The index entries are sorted by POC per GOP (to get display order in the DGI file), but it seems that it is not implementing the GOP boundaries correctly in this case, so the negative POC values in later GOPs migrate to the first GOP.
Actually, have an older one and the kitchen was from the early 70s. You can imagine. Homely country look with goofy masonite wainscoting and rose-adorned wallpaper -- projectile vomit -- people used to like this stuff?! So I gutted it to the bare walls and plywood subfloor. Thomasville cabinets (functional not foo-foo) (Made in USA) and quartz countertop are in. Backsplash subway tiling in progress. Saving thousands doing it myself but didn't plan on massive hand cramps. Final result is gonna be sweet!
I'll try to find time to look at your issue tomorrow.
SPOILER: My avatar now faces the other way (toward the post instead of away). It feels more welcoming and inclusive.
Wow that's a lot of work. Didn't notice your avatar change though (looks pretty much the same so...)
Good to know the bug is fixed. I'm gonna test it against my other samples to make sure it's working perfect. (Hint: there was some seeking accuracy problem I noticed, so I'll double check.) Will let you know!
Definitely negative POC shifting problem is solved.
Seeking inaccuracy still presents. I have uploaded a minimized sample to your FTP "259test4.hevc", please check frame 24, sequential and random access.
It's not the negativity; it's the fact that the leading RADLs are B frames, and leading B frame code hasn't been implemented for HEVC hithertoo. This is actually the first HEVC stream seen here that has leading B frames. I knew the day must come, so now it's time to take care of business.