Indie Film Live

Tuesday, November 01, 2005

Indie Film Live

Hi guys,

Here is an update with some questions. Today we shot a replica of an interior car shot from a recent job which utilized 2k dpx files canned from telecine. The shot framed the driver of a car with a green screen behind. Both windows were open (we should have shot one with windows closed as wel). The object of the test was to determine the effectiveness of the keying ability of both the codec and the HD100e.

three files were produced.

1. footage shot to tape using mpeg2 hdv @ 30p.
2. footage shot to disk uncompressed @ 60p
3. footage shot to cineform large format @ 60p.

We then took the footage through after fx into cine (dpx) files which were keyed in combustion. We havn't tested the mpeg2 file yet but the other two keyed with different degrees of success. The uncompressed was perfect while the cineform required more work but although it wasn't AS clean as the uncompressed (8bit? see later on) file the result looked great. The biggest probelm was the jagged edgeing we got which we had to correct for using matte blurring etc. . I expected some artifacts but wanted to know what you thought of the result and to see if there is any way to improve on this.

There was also another interesting result. When we converted to cin files we also exported 8bit tga's for comparisons. Now the uncompressed avi was captured using widows media capture and was fixed at 24bit so that is one thing we want to resolve for proper testing. But the downconverted cineform (from 10bit cineform to 8bit tga) exhibited what appeared to be identical offspring. ie no difference between original 10bit cineform and exported 8bit clone. We difference matted the two and found no difference at all. Perhaps we are missing something but we are very interested in understanding the best process through all of this to make sure we are getting the best possible result. We are using combustion in our work flow for various reasons but mostly because Glen can compare benchmarks against previous work in that software, which is helpful.

The camera, convertor and process was very impressive over all. Everything worked BETTER than expected. We got very excited about it. There was even a litle dancing. We also shot with detail off - which is worth noting after our last discussion on the stretchies. No stretchies noticed yet on this clip. We also converted DPX files used on a feature film we just finished and we got the same jagged lines on the key. Aside from the jagged edge the same key (copied and pasted the key operator) worked - again bluring was used to enhance. Not ideal but definately workable in many cases. I will upload pictures of both dpx and hd100 footage tommorrow for your perusal. Let us know if you have any thoughts on this.

Simon Hansen

Atomic VFX


  • If Combustion is using the Video for Windows (VFW) version of the CineForm codec, it will be limited to 8 bits. That would explain the identical difference mattes. You might want to try the 8 bit tga export from After Effects (with the CineForm importer active). Then difference the 8 bit tga with the 10 bit CFHD AVI.

    By Blogger Jeff Youel, at 7:27 PM  

  • Hi Jeff, thanks for the response. We didn't use the cineform avi in combustion in the test. We only used cineform in After fx where we exported 2 copies of the file.

    1. 8bit tga
    2. 10bit cineon

    these two files were imported into 10bit work space in combusiton. When differnce mattes were applied the file exhibited no differences. Does that make any sense to you?


    By Blogger simon, at 4:28 AM  

  • That sounds like something I going to want to try here, as no it doesn't make sense. With After Effects the CineForm importer (yes it must be active) will decode 10-bit YUV to 15-bit RGB (AE is realy a 15 bit app not 16.) The exporter would then use this data to the required significance. There should be some small difference in the LSBs (least significant bits.)

    By Blogger David, at 7:49 AM  

  • Hi Guys,

    We uploaded a few files last week and I think I mis - posted it somehow. anyway here is the latest key test material. I have uploaded two files to
    Pasword and user as per normal.

    1. key_cineform.jpg
    2. key_cin.jpg

    Both files are digital clones of a film sequence scanned to dpx sequence in a datacine at 2048x1556. We have zoomed in to show the detailed edge of the two results we achieved and then saved that as a jpg.

    key_cin.jpg is the cineon clone created in AE and then keyed. (uncompressed)

    Key_cineform.jpg is the same sequence converted to cineform and then to cineon and then keyed.

    The edge difference is clear but in this case workable using additional composite tools. Our ideal is obviously not to have to work the edge to much and we expected some degradation of the original once compressed. Our question is whether this is the best possible result using a compressed image? Can we try compressing less (larger file still compressed). I remeber david newman mentioning larger file benefits max out at a certain point.

    let me know your thoughts,


    By Blogger simon, at 10:13 AM  

  • We have higher quality modes for 2K scans, particularly designed to deal with film grain, but it also helps in keying. When you convert your DPX files to CineForm Intermediate (note: I'm working on a batching tool to do this without AE), select the quality of "Film Scan 1" or "Film Scan 2", both our recent 2k project used "Film Scan 1". There is a mode above "Film Scan 2" called "Overkill", I haven't yet found a need for it.

    Another difference between the DPX test and CineForm workflow is 4:4:4 and 4:2:2, we don't have a 4:4:4 compression on the market yet. Some of the edging is a chroma resolution issue not a compression issue; that is why the horizontal edges are less defined than the vertical. As your project will work from 4:2:2 HD source there will be no compromise in using 4:2:2 based compression. Add for film scan projects we find the 4:2:2 10-bit plenty for the majority of indepent film's needs, but we plan to add 4:4:4 in the future (not soon.)

    With the upcoming Wafian recorder and new features within Prospect HD, we are in the process of adding new quality levels for green-screen/effects work -- just like the Film scan mode for 2k. Basically introducing levels above "Large" for higher performance systems that can handle this type of real-time compression. As systems increase in speed, the quality can be increased until (eventually) CineForm Intermediate will be mathematically identical to uncompressed.


    By Blogger David, at 11:52 AM  

Post a Comment

<< Home