Online Flash Video Transcoder

[Tropican]

Hey all,

I don't know where to post this or if its been posted before (searched it couldn't find it), but I found this nifty online flash video transcoder. It converts to DivX .avi and .mov. It does add a frame with credit info at the end, but that's easily removable.

http://digg.com/software/Convert_Youtube_Type_FLVs_to_DivX_MPEG_4_Online

Also it says "lossless" on the site but I am almost completely sure they are using a compressed domain transcoder based on the diagrams they have up there.

[foxyshadis]

No, it appears to be an honest lossless converter. h.263+ is mostly a subset of mpeg4, and sorensen spark doesn't even implement it all, so software to rewrite the bitstream for mpeg4 is quite possible. Of course, VP6/flv4 videos are right out. If the server ever stops returning busy messages I'll try it out.

Compressed domain transcoders are a lot harder to implement than they sound, anyway, and their usefulness marginal compared to a full decoder/encoder. (A bit more useful in the case of keyframes though, like jpeg or mjpeg, but not much.)

[Tropican]

No, it appears to be an honest lossless converter. h.263+ is mostly a subset of mpeg4, and sorensen spark doesn't even implement it all, so software to rewrite the bitstream for mpeg4 is quite possible. Of course, VP6/flv4 videos are right out. If the server ever stops returning busy messages I'll try it out.

Compressed domain transcoders are a lot harder to implement than they sound, anyway, and their usefulness marginal compared to a full decoder/encoder. (A bit more useful in the case of keyframes though, like jpeg or mjpeg, but not much.)

Interesting. Would it be possible to confirm this by doing a byte comparison on the raw streams from the source and resulting files?

[foxyshadis]

That wouldn't be possible due to the huffman input differing, probably, and checking after the first stage of decoding would be difficult. But you could check for identical output if the same decoder is used (like mplayer, ffdshow, or ffmpeg).

It has precedent in the divx3->mpeg4 conversion (http://forum.doom9.org/showthread.php?s=&threadid=85229) in ffmpeg, anyway; hopefully this will be added as a patch to ffmpeg and not kept proprietary.

Since I just got a conversion done, I'll test.

Results: Inconclusive. It seems to munge the timestamps just enough that comparison through avisynth is impossible, but it does show a very strong hint that it's yes: a number of frames are -Inf PSNR, most are -50, the rest are -30, so I still can't tell if it's dropping incompatible information or not. I'll try to extract each frame raw and compare that way, should save the problems with DirectShow.