"The Food Police", a 48 hour-like film
 

We've just wrapped up our 48-Hour practice film, "The Food Police". It's a comedy in the Detective/Cop genre.

Prop: A newspaper
Line of dialog: "That was rude"
Character: Burns or Bertha McDonnell, gastroenterologist

Other than "Julia" and "Mr. Roberts", we had no actors. Everyone else onscreen was part of the crew.

Filmed with a Canon 5D Mark II running the Magic Lantern firmware, using a Cinevate DSLR rig, Canon's 70-200 mm f/4L and a Sigma 20 mm f/1.8. The "dash cam" view was done with the 20mm just carefully placed on the dash. All of the audio was recorded in camera using a JuicedLink CX231 preamp (mgain=0 dB, dgain=+12 dB).

I posted it to both vimeo and youtube. Vimeo looks slightly desaturated compared to the original Quicktime file, but the comparison to youtube is pretty stark -- view the image in full for the full craptastic compression. In fairness, it has improved slightly once processing was finished, but vimeo is still far superior.

Now that youtube has finished all of its processing and the HD mode is available, the compression is nowhere near as bad as it was initially. "The Food Police" on youtube in HD compares fairly well to the SD vimeo version, but once it is in fullscreen mode it is pretty clearly only 560x340 rather than the full 720p that vimeo is using.

That's a fun video. There's nothing like the 48-hour deadline to focus people on a project. The "ticking clock" is an excellent device for increasing the tension in drama - and it works in real life as well!

I especially liked the wide lens in the car shots. How did you lock the camera down?

So, were there lessons learned that you plan to apply to the next film?

We didn't. It was just sitting on the dash and I begged the actress to take the turns easy. The mic (an AT 4053) was resting between the two seats.

1. Talent is not allowed to wear white. Her blouse blew out the exposure in every scene.
2. Check all XLR cables for good solder joints. The one built into the boom was crackling. Luckily I was monitoring the audio with Magic Lantern and caught it after the second take.
3. Don't forget to record room tone. Luckily I remembered to get it for the exteriors
4. White balance. White balance. White balance.
5. We can get by with a much smaller crew.
6. Be careful trying to mix very different lenses. The 20mm feels so much different than the 85mm that it seemed like a different movie. Even the 70mm range of a zoom lens feels different from the 200mm end.

Since this wasn't an official 48 Hour Film Project entry, the pressure was much, much lower. I went to sleep on Friday and Saturday night around midnight and we didn't start filming until ten on Saturday. This is in stark contrast to our previous entry when we had two all-nighters with a marathon 18 hour editing session.

For rendering / export, FCE can save a "self contained quicktime file" much faster (like 1/10th the time) than an "export via quicktime conversion". mpegstreamclip can render the Final Cut selfcontained file into h.264 in 1/20th the time of Final Cut. When minutes matter at the end of the time limit, these things count!

Nice job Trammel. Camera and ML looked good.

If I ever do another 48 hour, it will be with just a small ensemble. Best experience in making a film I ever had was with two actors and myself shooting. What happens on 48 hour shoots is you tend to gather a lot of people to cover contingencies, and then you don't want to dissapoint them and leave them. Your script has to be written with all in mind. With feeding, locations finding, and logistics it gets really involved. You lose a lot of time on that stuff because it becomes more like a party.

Not bad, especially for a 48 hour challenge (I've had a few of my DVC entries turn into that against my will), though I may have had the guy in the car throw the rest of his pastry at that woman. Or just shove the whole thing in his mouth and drive off.

A question, though:

May I ask what makes you say that? Everything in my experience shows that Youtube HD is encoded at 1280x720. The dimensions in the Embed field underneath a video's description are suggested values for the player interface someone might want to embed in a page on their website, but they can be customized according to an individual's needs, and they aren't the actual dimensions of the video.

The Youtube version of your film looked fine to me in the player, even sized to full screen it was as sharp as any 720 HD I've seen on Vimeo. I pulled the files themselves from my browser's cache, and as I expected they're both 1280x720. Comparing them side by side, as in these two attachments (the strands of hair in the first sample and the gravel in the second), I can't tell the difference.

I don't mean to be confrontational, don't get me wrong, it's only that I've heard similar claims before but never heard them explained.

I was basing that on the embed sizes and how it looked played full screen in YouTube HD. Your clips from the two videos sure look like they are the same resolution; my less scientific single-blind test with two subjects both agreed that the youtube video was lower res. However, all three of us could certainly be wrong.

I've also upload it to Exposure Room and OpenFilm (not yet available) for further comparisons. The vimeo link should have the option to download the original h.264 Quicktime if you want to compare that one, too.

Quick edit: You can grab an uncompressed TGA copy of the sample image I refer to here, and it should get rid of any lingering compression artifacts that might muck up the comparison. I shouldn't have bothered with a JPEG in the first place, sorry.

----

I figured there might be something to gain from this test, so I didn't bother waiting for the OpenFilm version; each of the four existing encodes of the film (the original MOV, the Vimeo-recompressed MP4, the ExposureRoom FLV, and the HD Youtube MP4) was opened in VirtualDub. I hit Ctrl+G to go to frame 3636, and by setting that as the In, and the next frame as the Out, I was able to save an image sequence of only one frame. VirtualDub's timeline selection works that way, marking the boundary between the current frame and the previous frame when you press either Mark In or Mark Out, so doing what I did let me select only frame 3636.

Each of these stills was an uncompressed Targa file that I then loaded into a 2560x1440 Paint Shop Pro X canvas and, by way of Snap to Grid, positioned them the way you see them. Threw on some numbers and exported the final image, JPG compression artifacts not manifesting themselves in any way harmful to the quality of the individual shots.

Maybe going through all of that was a bit much to prove the point, but I thought somebody might be interested in seeing a comparison.

For reference, the plugins necessary to open each file type in VirtualDub, none of the ones used here being natively supported, are Quicktime Import v0.2.0.0 (all the way at the bottom, though his QTSource up top can be useful for AVISynth users) for the MOV version, version 1.1.0 of the FLV Input Plugin (second down the list) for the ExposureRoom FLV, and a build of ffdshow tryouts from a few months ago to let a VfW application like VDub handle the H.264 streams.

Thanks, Robert. That's a great comparison example. Would you mind if I reused it elsewhere?

#3 seems to be the most faithful to the original, followed very closely by #2. The brightness in #2 seems to be too high, leading to some additional noise in the shaded part of the tire. #4 definitely has the compression set much higher -- the distinctive X designs have completely been obscured by MPEG noise. But I would agree that they are all the same resolution.

I'm guessing that the images are:
1. "Original" QT
2. Vimeo
3. Exposure Room
4. YouTube HD

However, when I playback the local 720p h.264 Quicktime file, I can easily read the lettering on the sidewall of the tire. None of the four that you have posted are that clear. I wonder if that is a vimeo thing or a VirtualDub Quicktime importer issue. Do you know if the crushed blacks that were fixed by Quicktime 7.6 were also fixed in your tool?

The answers are as follows:

1: Youtube
2: Original
3: Vimeo
4: Exposure Room

I took these stills one step at a time, being exceptionally careful to keep them in order, so as surprising as the results may be I can confirm--for what my word is worth to you, anyway--that I have the numbers associated correctly.

The highly visible blocking on number four is usually symptomatic of Flash video, and though Youtube compresses FLV versions of some content, even in HD, I'm not sure of the conditions that trigger it. In this instance it delivered an MP4 container with H.264 video inside.

As far as decoding the video is concerned, I should have mentioned before that the Quicktime plugin I'm using makes direct use of functions in the Quicktime API to decode videos, namely MoviesTask, DecodeSequence and ICMDecompress, which is the one recommended for H.264 files and is the one I explicitly set the plugin to use for this test.

The clarity issue may simply be the Mac/PC gamma difference rearing its ugly head; I'm using a Windows machine, and CRT calibrated as best I can to match sRGB, but I don't have a hardware puck for that, and it's not exactly a broadcast grade display. God only knows how many variables there are in this equation. The PNG you just attached is quite dark on my display, and while I can color correct it to make the sidewall text brighter, it's no more or less detailed than the images I posted. Likewise, comparing the original MOV, opened in Quicktime Player (Pro, with QT 7.6.2 installed) with the TGA file saved from VirtualDub, the detail level is equal. QT Player does render the shadows in H.264 files too brightly, as has been the case for a while now, but the information is there, and brightening up the Targa it matches the visual fidelity of the source movie file as displayed by the Windows version of Quicktime Player.

I'm no color management expert, and I don't know that there's a simple answer to making these things consistent across disparate machines. My intent was only to demonstrate the spatial resolution of the various compression techniques on the different sites, which I thought the gravel did rather well. I hadn't even considered the tires, to be honest, and the JPG artifacts do become noticeable if you brighten that up. I've taken the liberty of uploading an uncompressed TGA version of the file (scroll down past the ads to find the download link) that should eliminate the question of compression artifacts, and provide the highest quality possible.

Feel free to use the file wherever you like, if you think you can get some use out of it. I hope it helps!

It is finally available: Openfilm » Videos » The Food Police and is an "Editor's pick" right now.