Author Topic: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread  (Read 948047 times)

Offline Swatch

  • Full Member
  • **
  • Posts: 273
  • Official Aerospace Engineer as of June 13th, 2009
  • Cincinnati
    • ProjectApollo/NASSP: Virtual Systems and Flight Simulation of the Apollo Program
  • Liked: 28
  • Likes Given: 16
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1820 on: 06/29/2014 07:53 PM »
Or better yet... how many hours per MegaBlock... that seems to have been the defining unit of modification for this project. :)
Ex-Rocket Scientist in Training, now Rocket Scientist!
M-F trying to make the world of the future a smaller place through expanding horizons...

Offline Quialiss

  • Member
  • Posts: 75
  • Liked: 82
  • Likes Given: 29
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1821 on: 06/29/2014 08:00 PM »
Here's a fun estimate:  How long would it take to do this again?  Separating out the time it took for all of us to learn how to fix the video from how long we spent actually fixing it. 

One of the few things I do know how long it took me to do: 6 hours for 38 p frames (2 sets).  SwissCheese also went over them after me and fixed my mistakes, and then I went over them again after that to fix up DC issues, so lets bump that up to ~10 hours.

So ~10 hours, ~40 p frames

1 p frame ~ 15 minutes. 

13 parts * 19 p frames * 15 minutes = ~60 hours to fix all the p frames

The i frames are harder to account for because they took much longer to fix, and thus tended to be done in stages.  I can throw an estimate of ~2 hours per frame for fixing up the DC values after all the good macroblocks had been found.  The actual time I spent on this varied wildly per frame and was rather higher because I was learning how to do it, but I don't think redoing even the worst of them would would take me more than 3 hours of work now.

Anyone have any time estimates for how long it took to find the good data in the i frames? 

Offline mhenderson

  • Member
  • Posts: 69
  • USA
  • Liked: 101
  • Likes Given: 18
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1822 on: 06/29/2014 10:34 PM »
Here's a fun estimate:  How long would it take to do this again?   

Excellent assessment ... as we have come expect from you. This project has moved the art of garbled video restoration a large step forward.  I suspect that three or four key people -- knowing what we know now, and using the tools that were built -- could fix the same data stream in less than a week.

Offline princess

  • Member
  • Posts: 65
  • Liked: 106
  • Likes Given: 25
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1823 on: 06/30/2014 10:05 AM »
I suspect that three or four key people -- knowing what we know now, and using the tools that were built -- could fix the same data stream in less than a week.

With the TS-level tools and experience we have now, I reckon I could probably provide a fixed TS file in under a day. That would leave four days for everyone else to do the MMB fixing.

I just had a thought - if SpaceX decide to drop the next webcast, I'll provide an equivalent length TS file with correct timestamps and frame headers for everyone to get started on ;)

Online S.Paulissen

  • Full Member
  • ***
  • Posts: 393
  • Boston
  • Liked: 286
  • Likes Given: 440
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1824 on: 06/30/2014 02:46 PM »
I only say this half-jokingly.

You guys should think about making a company to do this for money.
"An expert is a person who has found out by his own painful experience all the mistakes that one can make in a very narrow field." -Niels Bohr
Poster previously known as Exclavion going by his real name now.

Offline Quialiss

  • Member
  • Posts: 75
  • Liked: 82
  • Likes Given: 29
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1825 on: 07/02/2014 06:28 AM »
To celebrate some awesome t-shirts arriving at my doorstep, an update on what I've been working on.  It's tangentially related to my coursework... so I tell myself.  Live problems are always so much more interesting.  :)

I made a minor breakthrough today in automatically finding good MB start positions, with my script only hitting two false positive matches before finding the manually located start position, down from 10+ false positives.  I'm working with frame 10 for a fairly simple start case to work with, but I've tested it on some other frames(and the end of frame 10) with similar results.

Current methodology:

Test cases run on macroblocks from the new start position to end of the row(arbitrary end point but effective enough for the moment)
lum/chrom levels.bit_length() == size.  The levels should fit exactly into the size, if they're smaller, the size/value pair is wrong.
lum size < 7 and chrom size < 4.  These are derived from analysis of repaired frames. 

Require 15 blocks from new start position decoded without errors.  This is largely a crutch to shore up the lack of better tests, as it catches DC clipping from bad lum values and alignment.  I put a high priority on being able to shrink this value without increasing false positives as it means smaller pieces of data will be recoverable.

I haven't tested how many false positives each of these cases catches, it would be interesting to check.  I know reducing the error free blocks to 10 causes ~70 false positives to show up. 


The automatically generated MMBs, with my notes on what to make new test cases for.  Cut off point is just before the next correctly found block.  Since it doesn't know which one of the candidates is correct, it just blindly continues on trying to find good start positions.

08:14:-2,9:14:56667 Texture and lum
08:14:-2,9:14:56677 Texture, lum and alignment
08:14:-2,9:14:56683 *Hand picked solution
08:14:-2,9:14:56699 Texture
08:14:-2,9:14:56711 Texture, alignment
08:14:-2,9:14:56727 Texture, alignment
08:14:-2,9:14:56744 lum, alignment

I know I can build a test case to rule out the ones marked with lum errors.  There are legitimate times where the lum values change that much in a block, however, the rows/columns of very different lum values are a dead giveaway that there's an error.  It'll just be a much more *complicated* test case than the ones I've already established.

Errors in the DCT causing the texture of the block to be bad... I would love to hear if anyone has ideas on how to test for those!

Realigning the chunks of data MIGHT not be too difficult.  At least in this one case, it's notable that if the data is out of alignment, dc clipping errors show up everywhere, and go away when the data is correctly aligned.  Nudging the position back and forth 1-5 blocks would work here, but that assumes it was close to correct in the first place.  When I get to this, I'll check to see if I can use the dc clipping errors to verify alignment in other places or if it's just showing up so strongly here because of the detail in the end of the rocket. 

I've been toying with the idea of advancing the mb position after testing the mean block size of bits(~100).  This would actually be horrible for the first error in frame 10, as the error is in a block of size ~400, and in any case would require putting the 'wiggle' alignment checks in to get any good results out of it. 


Amusing complications:  My unit tests for valid block values are now 'better' at detecting errors than the error log from ffmpeg.  This is a problem because I'm using the error log to decide when to start looking for new start positions! I correctly detect that the next error in frame 10 starts at 08:21, with a too high lum size, whereas the error log finds the next error at 13:21. 

Offline mvpel

  • Full Member
  • ****
  • Posts: 1124
  • New Hampshire
  • Liked: 1295
  • Likes Given: 1686
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1826 on: 07/02/2014 10:28 AM »
To celebrate some awesome t-shirts arriving at my doorstep, an update on what I've been working on.  It's tangentially related to my coursework... so I tell myself.  Live problems are always so much more interesting.  :)

I think Dr. Young will be quite interested to discuss this with you. He's been working on cognitive video quality analysis, and automated detection of visual artifacts in a decoded video stream, and at first glance it seems like that work might be able to automate the selection of the best fix from the list of candidate MMB positions which your code produces.
"Ugly programs are like ugly suspension bridges: they're much more liable to collapse than pretty ones, because the way humans (especially engineer-humans) perceive beauty is intimately related to our ability to process and understand complexity. A language that makes it hard to write elegant code makes it hard to write good code." - Eric S. Raymond

Offline princess

  • Member
  • Posts: 65
  • Liked: 106
  • Likes Given: 25
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1827 on: 07/02/2014 04:38 PM »
I just received a box via FedEx containing a SpaceX baseball cap and 5 SpaceX shirts, including a CRS-3 mission patch shirt! Yay!

Hopefully the SpaceX staff who arranged it are reading this thread - thank you very much for sending them, they're very much appreciated. You guys are the best!

Offline mvpel

  • Full Member
  • ****
  • Posts: 1124
  • New Hampshire
  • Liked: 1295
  • Likes Given: 1686
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1828 on: 07/02/2014 07:59 PM »
I just received a box via FedEx containing a SpaceX baseball cap and 5 SpaceX shirts, including a CRS-3 mission patch shirt! Yay!

The box made it to England already, but not to Massachusetts yet?  ??? Poor Linda's starting to find me annoying when I ask her if it arrived yet on a semi-daily basis. ;)
"Ugly programs are like ugly suspension bridges: they're much more liable to collapse than pretty ones, because the way humans (especially engineer-humans) perceive beauty is intimately related to our ability to process and understand complexity. A language that makes it hard to write elegant code makes it hard to write good code." - Eric S. Raymond

Offline princess

  • Member
  • Posts: 65
  • Liked: 106
  • Likes Given: 25
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1829 on: 07/02/2014 08:40 PM »
The box made it to England already, but not to Massachusetts yet?

Aww I'm sure it will come soon... I guess FedEx UK must just be very efficient ;)

While you're waiting, I've attached a pic of what I received, with the CRS-3 T-shirt taking pride of place in the middle.

Offline deruch

  • Senior Member
  • *****
  • Posts: 2218
  • California
  • Liked: 1729
  • Likes Given: 3594
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1830 on: 07/02/2014 08:46 PM »
I just received a box via FedEx containing a SpaceX baseball cap and 5 SpaceX shirts, including a CRS-3 mission patch shirt! Yay!

The box made it to England already, but not to Massachusetts yet?  ??? Poor Linda's starting to find me annoying when I ask her if it arrived yet on a semi-daily basis. ;)

Yeah, about that......IainCole did a quick recount after some voting anomalies were discovered and you actually got bumped to #16.  No one had the heart to tell you, sorry bro.   :'(

just kidding  ;D
« Last Edit: 07/02/2014 08:48 PM by deruch »
Shouldn't reality posts be in "Advanced concepts"?  --Nomadd

Offline mhenderson

  • Member
  • Posts: 69
  • USA
  • Liked: 101
  • Likes Given: 18
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1831 on: 07/03/2014 10:37 PM »
Selfie. Thank you SpaceX, NSF, and the entire team.
** Someone should fix this. I haven't got the first clue how to start.  ;)
« Last Edit: 07/04/2014 10:41 AM by mhenderson »

Offline Dudely

  • Full Member
  • ***
  • Posts: 312
  • Canada
  • Liked: 109
  • Likes Given: 92
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1832 on: 07/04/2014 02:25 PM »
Looks like a bitflip to me  ;).

Offline mvpel

  • Full Member
  • ****
  • Posts: 1124
  • New Hampshire
  • Liked: 1295
  • Likes Given: 1686
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1833 on: 07/04/2014 08:02 PM »
 
Yeah, about that......IainCole did a quick recount after some voting anomalies were discovered and you actually got bumped to #16.  No one had the heart to tell you, sorry bro.   :'(

just kidding  ;D

Or maybe...



 :o :D
"Ugly programs are like ugly suspension bridges: they're much more liable to collapse than pretty ones, because the way humans (especially engineer-humans) perceive beauty is intimately related to our ability to process and understand complexity. A language that makes it hard to write elegant code makes it hard to write good code." - Eric S. Raymond

Online Chris Bergin

Selfie. Thank you SpaceX, NSF, and the entire team.
** Someone should fix this. I haven't got the first clue how to start.  ;)

HAHA! I was literally about to stick that into MS Paint and do a rotate - pretty much the limit of my skills - until I finally got the gag! ;D

Offline boo

  • Member
  • Posts: 1
  • Tasmania
  • Liked: 2
  • Likes Given: 0
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1835 on: 07/05/2014 11:57 AM »
Selfie. Thank you SpaceX, NSF, and the entire team.
** Someone should fix this. I haven't got the first clue how to start.  ;)

Looks like an "I" frame to me ;)

Offline meekGee

  • Senior Member
  • *****
  • Posts: 8244
  • N. California
  • Liked: 4465
  • Likes Given: 870
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1836 on: 07/05/2014 03:46 PM »
Selfie. Thank you SpaceX, NSF, and the entire team.
** Someone should fix this. I haven't got the first clue how to start.  ;)

Looks like an "I" frame to me ;)

Ok, that's just too clever.

ABCD - Always Be Counting Down

Offline mvpel

  • Full Member
  • ****
  • Posts: 1124
  • New Hampshire
  • Liked: 1295
  • Likes Given: 1686
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1837 on: 07/07/2014 04:07 PM »
Yay!  ;D The box finally arrived! Well, it arrived last Thursday morning on the 3rd, but I got the call from shipping this morning.  ::) The cap is a classy grey one with the SpaceX logo, and the CRS-3 Mission Patch t-shirt of course, and a few nice V-neck logo t-shirts.
"Ugly programs are like ugly suspension bridges: they're much more liable to collapse than pretty ones, because the way humans (especially engineer-humans) perceive beauty is intimately related to our ability to process and understand complexity. A language that makes it hard to write elegant code makes it hard to write good code." - Eric S. Raymond

Online Chris Bergin

Great stuff, and this is now about to go through 390,000 views. Given the way people find or revisit threads, this will easily get over the 400,000 mark.

Absolutely astonishing how this all progressed and the way it was appreciated by SpaceX/Elon.
« Last Edit: 07/07/2014 09:01 PM by Chris Bergin »

Offline getitdoneinspace

  • Member
  • Posts: 69
  • Liked: 81
  • Likes Given: 82
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1839 on: 07/08/2014 08:16 PM »
I heard that Elon was asking when you guys will be done with the restoration of the CRS-3 first stage landing video in High Definition. And the audio wasn't yet up to Dolby surround sound quality, but it was close.  ;)

Chris - You may want to let SpaceX know that their link to the restoration video is pointing to a private video (just see static).

Take a look at the following URL: http://www.spacex.com/news/2014/04/29/first-stage-landing-video

This is the URL they are using for the restoration video on the above page: 
« Last Edit: 07/08/2014 08:20 PM by getitdoneinspace »

Tags: