How do you feel about improving your completed works or, if you are a consumer, about other people changing their films? For example, I read that Ridley Scott decided to have the sound design in Blade Runner updated for its 25th anniversary, citing technological assumptions (e.g., hydro-electric power sources) not being in tune with today’s world. I do not know if the article is true, but it would be disappointing if so. Blade Runner to me represents a projection of the future; a possible one… an interesting one. So what if it turned out differently? Besides, mistakes are fun to catch, so long as they do not distract.

I tweak my work all the time from festival to festival, but if I had a feature I would draw the line at theatrical release. After that, one more cut for DVD (the director’s cut) and move on to the next film.

I posed this question to Zeki Demirkubuz at this year’s Boston Turkish Film Festival and he said the notion of never completing a film made him restless. How do you feel?

Last month Eugene and I helped shoot a zombie film by Max Bowen called Survival Instinct, as gaffer and second cinematographer (me), and boom operator (Eugene). In between takes we took pictures. Here are two of mine:

Thanks to Eugene for the group picture; check out my hat:

For OTW I had some noisy indoor footage captured with an XL2 in 2:3:3:2 (aka 24P advanced) mode. Naturally I was looking for a way to clean it so I decided to use avisynth, since it has a wide array of good free noise reduction plug-ins. Since the state of the art is always changing, I spent some time looking through the archives of doom9 (where all the avisynth enthusiasts are). I discovered a great FFT-based denoising filter called FFT3DGPU ^[1]. As you have no doubt guessed, this is because this filter takes advantage of your GPU! I thought it would improve performance so I gave it a try. Apparently it is a port of a well-known filter called FFT3Dfilter, which performs noise reduction in the frequency domain. Most other filters are spatial, temporal, or both (spatio-temporal).

The results are excellent, but before I show them, let me go through the avisynth script, since there are some points of special interest to XL2 users. Normally one would simply use a script like this:
avisource("file")
fft3dgpu
However one can obtain better results by eliminating the extraneous telecined frames. Progressive frames allow for more effective spatial noise reduction. It also makes for smaller files when you save the filtered clip. The way to do that is with the SelectEvery command. As its name suggests, this command selectively retains frames. By selecting only the progressive frames, we can pull down the 60i clip to 24P, like so:
SelectEvery(5,0,1,3,4)
This instructs avisynth to drop every second frame out of a group of five, thereby reducing its frame rate from 59.94i=29.97P to 23.97P. You might have to use a different set of numbers after the five in order to select the one interlaced frame.

The FFT3Dgpu filter accepts many parameters, the most important of which is the sigma setting, which controls the strength of the noise reduction. It’s a matter of trial and error to see how much cleaning your clip requires. You want to ensure that your denoised clip looks natural beside clips it abuts, if they are part of the same scene. Do not aim to clean every last bit of noise.

My finished script looks like this:
avisource("file")
SelectEvery(5,0,1,3,4)
fft3dgpu(sigma=3)
coloryuv(levels="pc->tv")

The last line accounts for the fact that the lagarith codec does not encode using the ITU-R 601 color space. Without it, your clip will have clipped highlights and shadows. Then I set the video mode in Virtualdub to Fast Recompress to avoid a color space conversion (i.e., staying in YUY2 for DV) and writing a lagarith file also in YUY2.

Finally, I insert the clip in Vegas by adding another take to the existing clip so I can easily switch back and forth with. Of course, this requires you to have already trimmed the clip in Virtualdub to the correct position.

1. ↑1 Users who wish to stay in their compositing application (e.g., After Effects) may be interested in RevisionFX’s brand-new plug-in, DE:noise.

Quick navigation tip from an Adobe developer: You can navigate the animation timeline by keyboard if you set up a shortcut.

Go to Edit>Keyboard Shorcuts (Alt+Ctrl+Shift+K) to bring up the Keyboard Shortcuts and Menus dialog:

Guess what: you can’t use PgUp and PgDn as the shortcuts. The most intuitive pair of keys I could get it to accept were Ctrl+* and Ctrl+/ on the numeric keypad. I then mapped these two keys to my ShuttlePro’s jog dial.

Were they really expecting users to use the mouse every time they want to advance frame? The writer jokingly says “please request this as a feature for next release“. Maybe they should have consulted the After Effects team?

I thought I would test the new Photoshop’s rotoscoping abilities. On the whole I’m pretty impressed! I might not have to use After Effects at all. One problem is that Photoshop is not smart enough to separate fields with interlaced footage, so you get things like this:

Good luck trying to rotoscope anything out of that. The solution, as others have noted, is to separate the fields first. One way is to do it in After Effects, as explained above. What do you do if you do not have that? You can use open-source tools Virtualdub and avisynth, as I will explain here.

Another minor problem with using videos in Photoshop is that the Save For Web tool tries to read the entire video when you try to save a single frame. It took me ten minutes just to generate the screen cap above! Save For Web huffed and it puffed, only to give up in the end because there was “not enough memory”. I only asked it to save a single frame, and navigated to it on the time line, so what does it need memory for!? After that it only took a few seconds to copy the canvas on to a blank document and save it with Save For Web. I bet Adobe will use this bug to entice people to upgrade to the next version: “New Save For Web…now compatible with video!”

Back on the topic of separating fields for interlaced footage… First you have to create an Avisynth file that looks like this for an NTSC video:


avisource("file.avi")
SeparateFields()
BicubicResize(720,480)


If you are using SD NTSC, as we did, this will create a 720×480 59.94fps video. The last line in the script instructs avisynth to expand the video vertically so that we can re-use the DV codec, which requires the video to be 720×480. Otherwise we would be working on a squished 720×240 video.

Save this script to a file with an AVS extension, open it in Virtualdub, and export it to something Photoshop can understand. The help file says that Quicktime, MPEG-1, MPEG-2, MPEG-4, MOV, AVI, and FLV are supported. Since AVI is merely a video container, you would think that Photoshop could use DirectShow to open any codec, but this is not the case. My understanding is that Photoshop uses Quicktime to open its files, so you are limited to the subset of codecs compatible with Quicktime. Well what are they? After some experimentation, I discovered that AVI files using DV worked, while those using YUY2, lagarith, MSU, JPEG lossless did not.

Once you have opened your file in Virtualdub you need to make note of the mark in and mark out points in Vegas. Go to the Trimmer and change the ruler to use Absolute Frames so you can input them in Virtualdub. You will notice that Vegas uses the time code of the tape by default, while Adobe naturally starts from the beginning of the clip. To make Vegas reset the counter to the beginning of the clip you have to go into the clip’s properties and reset it yourself, as shown below:

Now you should be able to determine the mark in and mark out points easily from the trimmer:

Go to Virtualdub and use Jump to Frame (Ctrl+G) to set the mark in-out points. Since we separated the fields in an earlier step, we need to multiply the frame numbers from the previous step by two to reach the same point in time. To use our example, if the mark in was 151 and mark out was 245, look for frames 302 and 490 in Virtualdub:

Here we used Mainconcept’s DV codec since it is compatible with Photoshop’s video import filter. Note the AVS file in the applications title bar, and that the RGB 16…235 check boxes have been ticked in both the decoder and the encoder. The first (decoder) ensures that the video imported into Virtualdub looks like it does in Vegas. The second (encoder) ensures that the rotoscoped footage looks like it does in Virtualdub.

You should now have a de-interlaced file you can rotoscope in Photoshop.

Image sequences

For short clips, you can make life easier, and get the highest quality to boot, by using images sequences instead of videos. Just load the Avisynth script in Virtualdub and invoke the image output filter from File>Save Image Sequence…. Use the TARGA file format and enter the file output directory, otherwise Virtualdub will litter its own installation folder with images. Once you have generated the image sequence you can import it into Photoshop by using the Image Sequence tick box in the File Open dialog:

Your screen may look different if you use the OS dialog rather than Adobe’s (which allows you to see the metadata). The next step is to select the frame rate, which is actually the field rate in our case. For NTSC it is 59.94.

The final step is to set the pixel aspect ratio. Since we are working with fields, we have half the vertical resolution of a frame, therefore the PAR has to be halved. Therefore, set it to 4:9=0.4444 for full screen and 16:27=0.5926 for widescreen. It should now look like it used to before separating the fields.

Once you are finished go to File>Export>Render Video. If you export to another sequence of images, you can combine the fields using an avisynth script like this:

imagesource(”retouched%03d.png”,000,147,59.94)
assumefieldbased
weave

The first line indicates that the sequence of images to load has file names retouched000.png to retouched147.png.