In LA

The web browser on my boss’s PS3 crashed when trying to render a web page containing Adobe Flash. It then displayed the error message “Flash is a poopy head”. In answer to my mildly skeptical tone, he sent me this screenshot he’d taken:

So there you have it.

I thought this screenshot might amuse. Click the photo to see it at full-size. At the top of the image are photos that include Chow Yun Fat, cropped to just the automatically recognized face. Below a line are photos that Aperture 3 thinks may be the same person. To refine the computer’s idea of what his face looks like and tag more photos with the appropriate name I had to select all Chow Yun Fat photos from the bottom part of the screen and drag them to the top, above the line.

For more info on the facial recognition feature in Aperture 3, check here on Apple’s website.

I have about 16,000 photos in my archive, and though the process isn’t completely automatic, the facial recognition feature made it much more feasible to tag all the people in these photos. Even just the fact that the application can display an entire set of photos cropped and resized to only show faces alone would be a huge help, the fact that Aperture 3 makes decent guesses is a bonus.

Posted via web from Zach’s posterous

I’m still working my way through the iPhone application class Stanford put online, specifically the Winter 2010 CS193P course. I’d highly recommend the course for anyone wanting to reinforce their knowledge of object oriented programming concepts, regardless whether they plan to ever write an app for iPhone, iPad, or Mac.

Below is a screenshot of my version of assignment #3 running on my iPhone. Every control on that screen works as expected, it’s pretty neat. Most importantly, I finally understand how and why.

You can find the lecture slide PDFs and other materials on Stanford’s website, and video of each of the class lectures are available as a free podcast in “iTunesU”.

edit: I’m wrong about the resolution issue — see correction and link to a full-resolution 2729×733 180º panorama at the end of this post.

I just shot and stitched a panorama while waiting for a ride. Though the Autostitch iPhone app does a great job, I can’t help but wish it would save the resulting panoramas to the photo library at full resolution.
 
The screenshot of the app displaying a section of the finished panorama is at least twice the resolution inside the app as is the version it saves. If the iPhone APIs limit the resolution at which apps can export to the photo library, perhaps autostitch could save to its own db and offer some other method of export (email, built-in webserver a la ‘Brushes’, flickr export?).
 
Regardless this one limitation, ‘AutoStitch’ is a lot of fun, and is well worth its $1.99 price.

Posted via email from Zachary’s posterous

– Edit 2009-06-19 12:32am –
It appears that AutoStitch does save its panoramas at full resolution to the iPhone’s photo library, but the photo library doesn’t display the images at full resolution. Nor does the iPhone’s mail client send the photos at the original resolution, instead it scales them to be much smaller.

I used the ‘Multi-Photo’ email application, which can send multiple photos via email without resizing, to send this panorama to myself.¹ Click the thumbnail below if you’d like to view the panorama at its full 2729×733 resolution.

Click this thumbnail to see the full-res pano.

1. Were I home and not out traveling, I’m sure the full-res photo would have easily synced to my home computer’s iPhoto library over USB, but that’s not an option in the field. [↩]

A screenshot of the Final Cut Pro Browser and Background Processes windows showing the SmoothCam Analysis in progress. Click the image to see it at a legibly large size.

I ran around the South by Southwest festival last week shooting behind-the-scenes video of my cousin AJ and his band on my Canon HV20, stabilized with a Steadicam Merlin. I’m still learning to use the Merlin –it is a fantastically engineered piece of kit, but it takes a light touch to float it about, and it’s much easier to use for choreographed moves than for impromptu run-and-gun documentary-style shooting.

Final Cut Pro and Motion both inherited a feature from Shake called “SmoothCam”, which is a motion stabilization filter. SmoothCam analyzes movement within a clip and moves the frame around to compensate for high-frequency movement while preserving the lower-frequency moves –i.e. it tries to remove bumps and vibrations while preserving smooth, intentional camera movements. I ran a few tests on some of my relatively smooth clips, and SmoothCam seemed very good at doing exactly that.

When the SmoothCam filter is applied to a clip, that clip is put into a queue and Final Cut Pro begins to analyze the motion in the clip. Once FCP’s analysis concludes¹ , the SmoothCam filter will have data to use for its calculations and its effect will finally become visible. One can then adjust the SmoothCam settings and see the results instantaneously (with a computer of sufficient power).

I’m happy to note that it is possible to have Final Cut Pro analyze the motion of a batch of clips, and then later if the SmoothCam filter is applied to any of these clips its results can be seen without delay. The motion analysis data for each clip is written to disk alongside the clips themselves.

Since I may or may not want to use the SmoothCam filter on many of my clips, and I don’t plan to do any more editing on my computer tonight, I’ve set FCP to batch analyze the motion in all of my footage from the last week. To do so, I enabled the “SmoothCam” column in FCP’s Browser, selected all the clips I wanted analyzed, right-clicked in the column and chose “Run Analysis” from the clip’s contextual pop-up menu. Now the analysis is puttering away, and tomorrow I can play.

The things one can learn by glancing at the manual!

Technorati Tags: Apple, Batch Processing, FCP Tips & Tricks, filmmaking, Final Cut Pro, Garbanzo Beans, Image Stabilization, Macintosh, OS X, SmoothCam, software, vfx

1. The entire clip is analyzed, regardless whether only a portion of it is in use in a timeline. A clip’s In and Out points are ignored. The analysis can take a while, and I would strongly recommend converting any clips in a long-gop format such as HDV to Prores beforehand for markedly faster analysis. [↩]

I’m doing color work on a couple of spec commercials today. This screenshot catches me attempting to match the color and luminance of two consecutive shots. Unfortunately the position of the lights was changed when the camera was repositioned closer to the actor, so the shot’s intensity, direction, and the overall color balance changed between the medium-wide and the close-up. When these shots are edited together and played back, the transition doesn’t feel optimally smooth due to the fact that the images don’t match well enough for the viewer’s brain to completely buy into the idea that the action is continuous from one shot to the next. In this screenshot,
 
I’ve captured a still of the wide shot and have overlaid it on the left side of the close-up shot as a split-screen view, to both guage the difference by eye and to see it in the variety of scopes provided by the application, the most useful guage to me for this matching work being the ‘Parade’ visualization at top right.
 
In this instance I’ve already dropped the highlights in the close-up quite a bit. They’re still about 17% brighter than the highlights in the wide shot, but if I darken them further it really drains all the life from the shot. I’m inclined to not match the brightness of the actor’s face exactly from shot to shot but instead to split the difference and just bring down the close-up to the point that it doesn’t jolt the audience at the transition. It’s a subjective art, and I’m enjoying the process; Muddling my way through using my photographic and editorial instincts.

Posted via email from Zachary’s posterous