![]() |
|
|||||||
| ShowAnalyzer Beta Discussion related to the ShowAnalyzer Beta. |
![]() |
|
|
LinkBack (1) | Thread Tools | Display Modes |
|
||||
|
Most of the major issues and features are now done and I've started looking at the free version.
I have 2 options: DirectShow or not DirectShow. And realistically, the more I think about it, the less I like DS. It is just so... um... SLOW. On a 2.4Ghz machine, I get about 90 frames per second. 90 fps!! SA gets 700 fps on the same machine. It would be an incentive to go for the "cost" version, but some people just don't/won't/can't pay anything. I think they should still get a decent product. And 90 fps is not decent! And I would have to completely reattack the problem when I port SA to linux. With an already available DLL, I probably won't have to. Which led me to find an MPEGDecoder.dll at http://nic.dnsalias.com/ I'm trying to write a "bridge" dll. So far so good. When I get a working version, I'm going to release it to beta and the world... {evil Dr. Evil laugh}. Jere |
|
||||
|
Next weekend, Cristina (my wife) is going to Spain and Gladys (my daughter) is going to Ecuador. With them on vacation, my available spare time will go up (of course).
No earlier than next Sunday (the 19th) and no later than the following Sunday (the 26th), ShowAnalyzer will move to beta and a free version will be available. Since there will be an opportunity to try ShowAnalyzer without shelling out a dime (that is 0.10 US$), the "price" of the fast version will be going up. I haven't decided how much, but I will probably be influenced by any feedback I get on the subject. Jere |
|
||||
|
There are days when I wish I had a feature request so that I could take a break from the parts of SA that drive me nuts.
Since edge profiling doesn't work in Yurp, I have put a way to disable it entirely (and it was requested). This results in an incredible increase in speed. About 4x. That's great and all, but it still doesn't address the problem of Europes commercials. I only have a couple example shows from over there. I can do Enterprise perfectly, but the Simpsons is killing me. The problem is (as I see it) that some undesirable segments aren't commercials. Commercials have to fit in specific spots so tend to follow certain rules. This "news" segment is one minute and 18.367 seconds long with a logo in the upper right hand corner. After the news segment is the opening credits for the Simpsons. Which is (amazingly) one minute and 18 seconds exactly. The show portion has a higher volume, a higher scene change rate, everything points to it being more of a commercial. Is this an oddity? Can someone from the other side of the ocean confirm this happens regularly? On a different note: SA needs to be able to learn. I would (ideally) like it so that as you are watching a show, if a segment is improperly marked, you hit a button and the status of that block changes. Then, during idle time, SA will look at the mistakes it has made and learn from them. This brings to light that SA (and comskip) output continous segments. Why? Why not have the commercials individually marked? Then teaching SA would be SOOO much easier, because you could say exactly which portion is incorrect. That would involve a new output format. I am really unhappy with the .txt file. Maybe a SIF (Segment Identification File) or a VSC (Video Segment Classification) or a JTBD (Just To Be Different) extension. Maybe something like: Code:
0:00:00.000 0:01:14.344 Commercial 0:01:14.345 0:03:01.988 Show 0:03:01.989 0:03:32.900 Opening Credits 0:03:32.901 0:04:02.840 Commercial 0:04:02.841 0:04:18.001 Commercial ![]() Anyway, I guess I've rambled enough for tonight. Jere |
|
||||
|
The more I think about it, the more I like the idea of a sa.xml file that has a variety of information in it. Including the commercial locations, channel number, number of frames, length, etc.
TiltRod, I think you could get your information from that. Any opinions? Jere |
|
||||
|
Over the weekend, I recieved an example show from the UK. Now I understand why commercial detection over there is so problematic. The number of black/blank frames is MUCH less than here in the States.
To combat the problem, I've spent the last 4 days incorporating using scene changes as border frames. I got it so that the example show was detected perfectly. It was really impressive. I mean from a technical point of view. The commercials were all 30.000 seconds or 15.000 seconds long. I was floored. And since there are no logos in the commercials, commercial detection in England should be a piece of cake. For those that don't know, I keep every show that has been donated to me so that I can run "sanity checks" on every change that I make. I have shows from the US (obviously), Germany and the UK. These come from a wide range of recording devices, too. It takes about 2 hours of pure processing, but it's a lot of air time to check. Anyway, I started up my "sanity check" program (DirMon), and went away. I was floored by the poor results. I guess I expected there to be "anomalies", but nothing like what really happened. I had two choices: give up or figure it out. I chose to soldier on and work on it and have made some good progress, but everytime I think I have something releasable, another anaomaly pops up. ARGH!! So for those of you wondering where the latest release is, I'm working on it. It's just taking longer than expected. I do thing that this will help improve commercial detection for everyone in the long run so it's worth the wait. I hope. Jere |
|
||||
|
I got so increadibly frustrated working on the scene change stuff, that I made it an option and released. The problem seems to stem around finding the correct audio threshold.
Speaking of audio, I've been working with Nic's Evil MPA decoder and I think it is the best way, but I only get 1 out of ever 3 frames with valid data. Go figure. I think I'll be able to find a solution and release by Friday for everyone to play with. That means there will be TWO versions of SA. One free, one not. The only difference will be the decoder, but there will be some consequences of not using my decoder. First, CC extraction will not work. Anything that depends on closed captions won't work... at all. If anyone knows of a non-directshow dll that extracts closed captions, let me know and I'll take a look at it. Second, live detection will not be possible. Third, the file will need to be read three times. Once during the audio dll initialization, once to extract the audio, and a final time to analyze the video. The first two are hard drive dependent. I can slow down the second one so as not to overwhelm your system, but the first one is pretty much full speed ahead. So the question may become, why use the free version of SA instead of comskip? I think you'll get much better detection even without the closed captions. Especially for Brits. All the functionality of SA will exist in the free version, like the communication with Sage, that allows for better tuning. It will just be slower and can't see the captions. Jere |
|
||||
|
It's a slow week for coding. A big inspection at work is consuming much more time than should be necessary. Oh well.
I received a request for an API to use ShowAnalyzer programatically. Upon some initial doubts, I now think it would be a good idea. So I'm rewritting SA's "infrastructure" so that there will be three files installed: SAConfig.exe, ShowAnalzyer.exe and ShowAnalyzer.dll. I'm combining the free and pay versions so there will be only one download. After entering a key in SAConfig, my decoder would be activated with all the appropriate advantages. This might open SA to piracy, but oh well. That's the plan. Jere |
![]() |
| Bookmarks |
| Thread Tools | |
| Display Modes | |
|
|
LinkBacks (?)
LinkBack to this Thread: http://www.dragonglobal.biz/forums/showanalyzer-beta/4-blog.html
|
||||
| Posted By | For | Type | Date | |
| ShowAnalyzer Beta - Dragon Global Forums | This thread | Refback | 04-21-2008 07:35 PM | |