Jorge Stolfi said:
You must be aware that MPEG and JPEG compression add very complicated artifacts to the frames. Roughly, each 8x8 pixel block is "simplified" independently of the other blocks. So one gets blurring and/or spurious streaks within each block, and spurious sharp edges between the blocks. For MPEG there are also complicated tricks to simplify apparent motion, that generate more artifacts when you try to grab a single frame. The video was recorded at high compression (= low quality) so these artifacts are quite strong. To make things worse, an edge-enhancement filter was used before compression --- which adds its own artifacts, such as spurious "grains" and "echoes" along sharp light/dark boundaries. Unfortunately, most of this damage is irreparable, since no filter can restore the information that was lost in the encoding.
For example, although the reddish flash spans a dozen pixels across, the video does not record any detail about its position, shape, and texture. The flash basically got simplified to one or two very bright 8x8 squares, with no internal detail; and that is all we are ever going to get.
What seems to be "the roof caving in" could also be a ring of black dust being blown horizontally al around the building, along the edge of the tar/cement roof, as the latter begins to be lifted off the underlying metal framework. This ring of dust becomes wider in the succeeding frames, covering more and more of the wall, from the top down.
I believe the rising column and mushroom is just a late convection artifact (i.e. not thrown out by the explosion, but lifted by buoyancy.) Like the mushroom of a nuclear bomb, or of a large gas tank explosion.
'tiz true -- resolution won't improve, and compression sucks detail out of the image. I am much more in tune with the effects of lossy compression in still photos and high resolution digital gray-scale images than I am with video that is low resolution stuff from the start, but I suggest that if not the detail, then the overall general motion and time frame of the motion are preserved in the context of a frame to frame analysis at 1/30 rate.
Also, compression artifact isn't what turns the initial white puff into red. Only heat does that. And though shadow and compression artifact play hell with the fine details of what is in the vertical smoke plume, there is very little doubt that a whole pile of garbage is falling out of it, streaking ground-ward in later frames.
Convection may cause billowing mushroom clouds and waft smoke and dust and water vapors aloft -- some of that is certainly going on; however I, for one, doubt rising heat alone will casually toss large, heavy chunks of debris almost straight and then quickly drop them on and around the building. It takes significant mechanical force and the impulse of a blast to accomplish that.
As for the "imploding" roof effect, I suggest (agree with you that) it seems equally plausible, and perhaps even likely, that the apparent downward roof movement is neither compression artifact nor implosion, but perhaps instead the visual effect of the sides of the top floor being blown outward and down, surrounded by dark smoke. When examined very carefully, there almost seem to be some horizontal contrails coming from that blast, too.
All great fun to sift through and speculate over. It is going to be even more interesting if I can get the audio track, pitch adjusted for slow motion, synched with the video. There are a bunch of very weird mechanical noises coming from that clip, too.
PS: I tried as best I could to hold to on-screen resolution when I grabbed the individual frames so as not to degrade the image further. I confess that I not too familiar with the .png format or loss of detail from a screen grab in .png format being re-displayed at the same on screen resolution. Side by side with the original video on screen, the reproduction seems true.
PPS: wiki says .png is lossless, bitmapped
