Barry_Green
Moderator
Okay, I looked and looked for where to put this, and I can't find anyplace that's logical for it, so... it'll go here unless someone suggests a better home.
I've seen questions about chroma keying with the new cameras, and how they compare to the older generation, etc. I've seen people keying a bald guy on an SLR and saying "hey, it keys great!" Er... not exactly a rigorous test, y'know?
So I decided to make a hellacious torture test for keying, and shoot it on a bunch of different cameras, and post the camera-original raw footage so you guys can try keying this stuff.
It's nasty. Some white lacey fabric and some red lacey fabric (what the heck is a "vault"? I dunno, but these fabrics called themselves "vaults" and they were $2 each at the fabric store.) Anyway, those lacey fabrics up against a green screen, and being blown around by a hair dryer. A red-haired woman, holding a cat, and her hair being blown around by another hair dryer. She's wearing a fuzzy-ish sweater, with a lace-ish scarf and a weird super-colorful scarf-ish thing. It's pretty much the worst-case scenario for keying you'll likely run into. I wish I would have had her holding a glass of water so you could try keying through that too, but I didn't think of it.
The contenders:
7D
GH1
HMC150
HMC40
HPX170
The camera-original footage is here: http://www.mediafire.com/download.php?nkmimwteckn
Be aware that it's about 124 megabytes. Each camera's clip is pretty short, only a few seconds. I wanted you guys to have motion footage and not just a lame still-frame, because the effectiveness of your keying is revealed in the motion.
Now, PLEASE HEAR THIS:
I didn't attempt to match the cameras to look the same; I just set them all to mild default settings, flat gammas with lower detail levels. Cine-D on the AG-cams, "neutral" on the 7D, "smooth" on the GH1. And I tried to get a proper white balance with each. With that said, they don't look the same. I wasn't trying to make them look the same. I didn't try to equalize out the color saturation or the toe or knee or anything. If you need them to look the same, that's another whole rigamarole, which is a lot of hassle and I didn't feel like doing it.
So, it is what it is. It should still be a valid test anyway.
Second: If you want to use the GH1 or HPX170 footage, YOU WILL HAVE TO STRIP OUT THE PULLDOWN. HPX170 was 1080/24pA, so pull out the advanced pulldown. GH1 has, of course, 2:3 pulldown in its 1080/60i data stream. If I see *anyone* posting clips with interlaced lines in 'em, I'm going to have to ban them for not reading.
Third: When working with the GH1 footage, you MUST use a pulldown-removal method that properly restores the chroma. Cineform NeoScene does so; I don't know what else does, but if you don't restore the chroma properly you'll be severely and unfairly handicapping the GH1 by throwing away half its chroma information.
All the footage is stored in complete folder directories as from the cards, so any editor that pukes without the full directory structure (and by that, I mean FCP) should be able to work with the files.
Feel free to garbage-matte the edges, some of the cams might be showing the edge of a door frame or the edge of a light fixture, so a garbage matte is a perfectly acceptable technique.
So -- here you go. If you can pull a successful key from any of these, you're better at it than I am. But if any skilled keyer would please take the time to pull the very best they can from this footage, from each camera, that would be awesome: it would be illustrative for us to see what you can get out of these cameras and how they stack up against each other.
More boring details: all five cameras were set up at once, all pointing at the same scene, using a preposterous combination of tripods, 1/4-20 to hot-shoe adapters, light stands as tripods, etc. They were all a long way from the scene and zoomed in, trying to minimize the parallax and get at least a reasonable approximation of the same direction of lighting on the subjects. I made an effort to make the relative magnification all approximately the same; the successfulness of that venture is for you to decide. The SLRs were at f/5.6 on their stock lenses, and I had to go to ISO 640 to get the appropriate brightness out of them. Exposure was set by piping the signal out to my BT-LH80 and using its waveform monitor. All sorts of user error could have crept in at all sorts of places, it's incredibly difficult to coordinate five different cameras and even remember how they all work, much less get every detail exactly right. I just hope they're at least in focus. If they're in focus with a decent white balance, I'll chalk that up as a "win." All cameras were set to 1080/24, all had detail turned down a couple of notches from default, except for the 7D which had the "neutral" profile which defaults to detail being zero anyway. If you want to add sharpening after the fact, go ahead. Light was such that the HMC150/HPX170 were at f/4 and 0dB of gain. The HMC40, unfortunately, had to have 6dB of gain on it; it just wasn't sensitive enough to go with zero. That's a big disadvantage for it, so if it manages to come out of this in any way respectably, it should only get better when lit appropriately for 0dB. Don't ask for me to test more cameras; if you didn't see it represented here, it's because I don't have one.
I've seen questions about chroma keying with the new cameras, and how they compare to the older generation, etc. I've seen people keying a bald guy on an SLR and saying "hey, it keys great!" Er... not exactly a rigorous test, y'know?
So I decided to make a hellacious torture test for keying, and shoot it on a bunch of different cameras, and post the camera-original raw footage so you guys can try keying this stuff.
It's nasty. Some white lacey fabric and some red lacey fabric (what the heck is a "vault"? I dunno, but these fabrics called themselves "vaults" and they were $2 each at the fabric store.) Anyway, those lacey fabrics up against a green screen, and being blown around by a hair dryer. A red-haired woman, holding a cat, and her hair being blown around by another hair dryer. She's wearing a fuzzy-ish sweater, with a lace-ish scarf and a weird super-colorful scarf-ish thing. It's pretty much the worst-case scenario for keying you'll likely run into. I wish I would have had her holding a glass of water so you could try keying through that too, but I didn't think of it.
The contenders:
7D
GH1
HMC150
HMC40
HPX170
The camera-original footage is here: http://www.mediafire.com/download.php?nkmimwteckn
Be aware that it's about 124 megabytes. Each camera's clip is pretty short, only a few seconds. I wanted you guys to have motion footage and not just a lame still-frame, because the effectiveness of your keying is revealed in the motion.
Now, PLEASE HEAR THIS:
I didn't attempt to match the cameras to look the same; I just set them all to mild default settings, flat gammas with lower detail levels. Cine-D on the AG-cams, "neutral" on the 7D, "smooth" on the GH1. And I tried to get a proper white balance with each. With that said, they don't look the same. I wasn't trying to make them look the same. I didn't try to equalize out the color saturation or the toe or knee or anything. If you need them to look the same, that's another whole rigamarole, which is a lot of hassle and I didn't feel like doing it.
So, it is what it is. It should still be a valid test anyway.Second: If you want to use the GH1 or HPX170 footage, YOU WILL HAVE TO STRIP OUT THE PULLDOWN. HPX170 was 1080/24pA, so pull out the advanced pulldown. GH1 has, of course, 2:3 pulldown in its 1080/60i data stream. If I see *anyone* posting clips with interlaced lines in 'em, I'm going to have to ban them for not reading.
Third: When working with the GH1 footage, you MUST use a pulldown-removal method that properly restores the chroma. Cineform NeoScene does so; I don't know what else does, but if you don't restore the chroma properly you'll be severely and unfairly handicapping the GH1 by throwing away half its chroma information.
All the footage is stored in complete folder directories as from the cards, so any editor that pukes without the full directory structure (and by that, I mean FCP) should be able to work with the files.
Feel free to garbage-matte the edges, some of the cams might be showing the edge of a door frame or the edge of a light fixture, so a garbage matte is a perfectly acceptable technique.
So -- here you go. If you can pull a successful key from any of these, you're better at it than I am. But if any skilled keyer would please take the time to pull the very best they can from this footage, from each camera, that would be awesome: it would be illustrative for us to see what you can get out of these cameras and how they stack up against each other.
More boring details: all five cameras were set up at once, all pointing at the same scene, using a preposterous combination of tripods, 1/4-20 to hot-shoe adapters, light stands as tripods, etc. They were all a long way from the scene and zoomed in, trying to minimize the parallax and get at least a reasonable approximation of the same direction of lighting on the subjects. I made an effort to make the relative magnification all approximately the same; the successfulness of that venture is for you to decide. The SLRs were at f/5.6 on their stock lenses, and I had to go to ISO 640 to get the appropriate brightness out of them. Exposure was set by piping the signal out to my BT-LH80 and using its waveform monitor. All sorts of user error could have crept in at all sorts of places, it's incredibly difficult to coordinate five different cameras and even remember how they all work, much less get every detail exactly right. I just hope they're at least in focus. If they're in focus with a decent white balance, I'll chalk that up as a "win." All cameras were set to 1080/24, all had detail turned down a couple of notches from default, except for the 7D which had the "neutral" profile which defaults to detail being zero anyway. If you want to add sharpening after the fact, go ahead. Light was such that the HMC150/HPX170 were at f/4 and 0dB of gain. The HMC40, unfortunately, had to have 6dB of gain on it; it just wasn't sensitive enough to go with zero. That's a big disadvantage for it, so if it manages to come out of this in any way respectably, it should only get better when lit appropriately for 0dB. Don't ask for me to test more cameras; if you didn't see it represented here, it's because I don't have one.
Last edited:
