The champions of long shots (1)

Know what ASL is? Keep up with the acronyms!

ASL (Average Shot Length) is a parameter that has been used by Barry Salt (Film Style and Technology: History and Analysis) and David Bordwell (The Classical Hollywood Cinema) to account for film style changes in history.

Sounds suspicious? Check out cinemetrics. There is a list of ASLs submitted by users of the program.

Unfortunately the list is far from comprehensive, not to mention complete (I am almost sure that they keep a private database for all that information…).

Anyways, just a glimpse of some of the top ASLs as shown in this list:

	Title	year	director	country	submitter	ASL
13	Andrey Rublyov ("Strasti po Andreiu")	1966 / 1971	Andrei Tarkovsky	Soviet Union	Grigori Utgof	34.1
12	Carmen Jones	1954	Otto Preminger	USA	Barry Salt	34.2
11	Opéra de quat'sous, L' ("La fiancee du pirate")	1931	Georg Wilhelm Pabst	Germany	Charles O'Brien	43.9
10	Voyage dans la lune, Le	1902	Georges Méliès	France	Torey Liepa	46
9	Dreigroschenoper, Die ("Seerauber Jenny")	1931	Georg Wilhelm Pabst	Germany	Charles O'Brien	47.2
8	Citizen Kane. Colorado boarding house sequence	1941	Orson Welles	USA	Yuri Tsivian	47.4
7	Great Train Robbery, The	1903	Edwin S. Porter	USA	Yuri Tsivian	49
6	Opéra de quat'sous, L' ["Chant d'amour"]	1931	Georg Wilhelm Pabst	Germany	Charles O'Brien	49
5	Dreigroschenoper, Die ["Liebeslied"]	1931	Georg Wilhelm Pabst	Germany	Charles O'Brien	49.7
4	Signora senza camelie, La	1953	Michelangelo Antonioni	Italy	Yuri Tsivian	55.6
4	Signora senza camelie, La	1953	Michelangelo Antonioni	Italy	Barry Salt	56.4
3	World (Shijie), The	2004	Zhang Ke Jia	China	Joshua Neves	57
2	Platform (Zhantai)	2000	Zhang Ke Jia	Hong Kong	Jonah Horwitz	67.7
1	Gertrud	1964	Carl Theodor Dreyer	Denmark	pm	82.4

I must say I am really surprised to see Gertrud has such a high ASL and Pabst appears four times on the list. But where is Hou Xiaoxian? Tsai Mingliang? Jansco? Angelopoulos? Tarr? Mizoguchi?

For the specific case of HHH, in this list only Three Times is given (29.5). But let me quote from James Udden’s Ph.D dissertation, HOU HSIAO-HSIEN AND THE AESTHETICS OF HISTORICAL EXPERIENCE:

The Development of Hou's Long Take Style

Film Title	Avg. Shot Length	% of shots with Camera Movement*	(% of shots with only slight reframing)
The Sandwich Man (1983)	17 seconds	26%	(7%)
The Boys from Fengkui (1984)	19 seconds	45 %	(16%)
Summer at Grandpa 's (1984)	18 seconds	32 %	(1 2%)
A Time to Live, A Time to Die (1985)	24 seconds	23 %	(8%)
Dust in the Wind (1986)	35 seconds	18 %	(9%)
Daughter of the Nile (1987)	28 seconds	11 %	(3%)
City of Sadness (1989)	42 seconds	29%	(14%)
The Puppetmaster (1993)	83 seconds	29%	(15%)

* This figure includes those noted at right

We see that The Puppetmaster has 83 ASL, which is slightly more than Gertrud. Does it make Hou the champion of long shots? I don’t necessarily think so. I don’t yet have all the data, but I believe the championship goes to Sokurov, Russian Ark, of which the ASL is—we proudly present—5400!

But wait a minute, there is a huge difference between Hou and those eastern European guys (Sokurov, Jansco, Angelopoulos, Tarr). They may have a long ASL, but their camera is always moving! What Udden has initiated in his study of Hou is important, yet remain to be refined.

First of all, Udden does not differentiate the kind of camera movement. It is perhaps a natural simplification for Hou’s case, but in doing so, we give up the opportunity to compare this set of statistics to everybody else.

Also, exactly because Hou does not use a lot of camera movement, the few of their instances are all the more important to show what kind of preference he has in cases of movement.

Interestingly, Udden does distinguish the category of “slight movement” to “entirely fixed shot”, which is perhaps not absolutely necessary. Barry Salt provides a good reason for not doing so:

I handle the objective treatment of camera movement by tabulating the numbers of shots with the different kinds of camera movement in each film, and again normalizing to the number that would be expected if the film in question contained 500 shots. The categories I use are Panning, Tilting, Panning and Tilting simultaneously, Tracking both without and with panning movements, movement involving the use of a camera crane, and Zooming. I have divided this last category into Zooming straight in or out, and Zooming with panning and/or tilting in my treatment of television shows. Only panning or tilting movements of more than 10 degrees are counted, as small movements to keep the actors well framed as they change their position slightly are made automatically by camera operators, and in general need no special thought about their relation to the director's ideas of staging. The same applies to small movements of a foot or so in the position of the rolling camera pedestal or dolly during the shot, and also of the height of the camera. I do not distinguish the different methods of supporting the camera, so that hand-held tracking and Steadicam tracking go in together with the traditional tracking with the camera on a dolly, or rolling pedestal in the case of TV.
*I underline.

In fact, I was a bit surprised when I first read that in The Boys from Fengkui the camera moves as much as 45%. My impression of the film tells a lesser number. Then I noticed that there is 16% of shots of imperceptible camera movements. What are “imperceptible camera movements”? In Hou, it is the kind of shot that you would only notice the movement if you are looking at the corners. And when I deduct 16% out of 45%, it gives 29%. In other words, the percentage of shot amounts to 71% in this film—more likely.

(TBC)

All that the industry is not doing right now but may be the essential characteristic of a future computer

Data Processing should not be separated from data storage. Memory should be integrated into the kernel. Would you imagine human memory to be a separate unit from the thinking part?

Data processing should be parallel. Instead of raising the clock frequency of execution, a pattern is to be fed to the processing unit and then generate a responsive pattern and revise the memory. This is the ultimate solution for energy efficient computing.

Graphics hardware is one of the most rapid growing parts in today’s computer industry. Yet while better images are being displayed, the capacity to recognize images has not improved at all.

Even if visual and audio inputs are used in computer, it never occurs to the industry that these kinds of inputs need to be multiple and pre-processed to be useful.

The way the computer relies on booting up the operating system to carry the most elementary functions is wrong. Computer should be like dogs, constantly on the alert. Any human voice and movement should be able to elevate its sleeping status to full functioning.

Visual and audio inputs are not to be treated separately. They are to be always jointly processed to improve the quality of recognition. For example, human mouth is a point sound source. If we can locate the orientation of the point, background noises can be more effectively eliminated. Also, the computer needs to understand that the person is only speaking when the mouth moves.

Visual and audio input needs to be multiplied and mobile (pan and pivot) to cover the whole space, not just a limited scope directly in front of the screen. A typical design would be two camera inputs at the two extreme sides of the screen (instead of the one-in- the-center design). As for microphones, at least three are needed: one in the center, for clarity, two almost on the same locations as the camera, open to sounds from even behind the screen. Together they will be able to track the movement of the person and zoom and refocus to always get a good picture of the face.

Face recognition should be the login and password for everything. This identification process may be joined by voice identification (a higher security level) and if needed, finger print and eye scan, etc.

Computer can be the majordomo that controls all electronic equipments in the house, light, room temperature, television and music related gadgets. But it needs to provide a safe and easy way to be overridden.

Multiple camera inputs arranged in a predetermined fashion will render a three dimensional face—what the face really is—and thus facilitate precise expression recognition.

Facial expressions are to be learnt. The learning process never stops. Yet the depth that the computer goes into distinguishing the facial expressions varies according to the exposure frequency—strangers’ faces will be remembered, but their expressions are not examined as closely as those of the frequent user (owner).