[mythtv] Commflag samples?
mythtv at hbuus.com
Tue Oct 28 04:49:07 UTC 2014
On 10/27/2014 4:02 AM, Mark Spieth wrote:
> On 27/10/2014 4:38 PM, faginbagin wrote:
>> It's not the first time the topic has come up, and there has even
>> been at least one patch submitted that attempts to fix it by
>> disabling the lowres mode introduced in 2010. Of course, disabling
>> lowres decoding would come at a very high price in terms of
>> performance. I think I've figured out why lowres decoding broke
>> commflagging of SD recordings and have figured out how to fix it
>> without losing the performance boost of lowres decoding. The
>> details can be found in this ticket:
> have you checked out
> I also generate log files and then process them with perl and R which
> provides a good overview of the variables over time as a graph to see
> what the features look like.
> There was also another issue with incorrect striding generating bad
> (re skewed) frames which means it wont work.
> I have also played a bit with a better logo detection algo. Nothing
> really good though. Dont have enough time to spend on it
> HTH or is useful. mark
I did skim your ticket, but didn't examine or apply any of the patches. Your work seemed to add significant new features. I also wondered why it's been sitting there for two years with only the command line method parsing accepted into the main line code. So I opted to first learn how it works in 0.27 and see if I could do something less substantive to improve accuracy.
I think I fixed what you call "incorrect striding". It's problem 1) I describe in my ticket. Or do I misunderstand?
Your discussion about logging and plotting did encourage me to set up a reasonably rigorous set of test recordings and test cases, so I could measure the impact of my changes. It's not well organized, but I'm confidant in my data. If it's of interest, I could take some time to organize and document the process of testing that I used, so others could do similar testing against their own set of recordings. At the very least, if anyone is interested in my summary spreadsheet, I'd be happy to share it, although it may take some explaining.
More information about the mythtv-dev