We've put together some demonstrations of how the Time Shaper software can work to change the resulting look of motion. Because we're dealing with very specific frame rate and motion representation issues, the very best way to see the results is to download the clips and play them on a system capable of the correct native display frame rates.
We filmed this aerobatic routine with a GoPro Hero 4 Black running at 120 frames-per-second. To achieve the necessary 360-degree shutter, an ND filter was used to shift the exposure of the camera.
This clip is rendered at 24 frames-per-second using the Tessive Osmium shutter throughout. In one scene, a speed ramp was used in Time Shaper to slow down the action and then smoothly ramp up to normal speed playback. In that sequence, the speed factor (from the burn-in) is displayed to show the ramp.
The resulting motion is very unlike normal GoPro footage. It's more cinematic and has very little judder even at the low presentation frame rate. This kind of footage could be used in conjunction with output from other cameras in a longer sequence and would match the feel of motion.
For comparison, we rendered the clips from the same 120fps input to 60fps using the Tessive Beryllium shutter, using the same edits as the 24fps version. This version shows the effectiveness of Time Shaper even at a 2:1 frame oversample, and allows use of the higher frame rate while substantially reducing judder.
ARRI Alexa Tests
These clips were acquired at 120 frames per second with an ARRI Alexa XT set to a 356º shutter. Some of the input clips were acquired in ARRIRAW, and some were acquired in ProRes format. The same input clips were used to create the outputs in both cases.
Time Shaper works best when the input clip frame rate is an integer multiple of the desired output frame rate. In this case, 120 fps is a factor of 5 above 24 fps, and a factor of 2 above 60 fps. It is possible to target other frame rates at non-integer scaling, but there will be frame-slipping on the input to match the realtime playback, or else the output will have an under or over-cranked look.
24 Frame Per Second Output
The 24 frame per second demonstration clip is a 450 MB h.264 file (or you can watch the embedded Vimeo clip, although the download provides better quality). There are two scenes in the clip, one with a vehicle driving past, and one showing human motion.
In each case, we show the traditional square-wave motion, followed by Time Shaper created output with varying synthetic shutters.
At 24 frames per second, we can substantially eliminate "judder" (sampling aliasing), but "strobing" (reconstruction aliasing) is still a property of the display or projector. Modern digital projection and most displays employ what would be termed a 360º reconstruction shutter. That is to say, each output frame is held on the display for a full 1/24th of a second, then an instant transition occurs to the next frame. Film projectors did not work this way, but instead had two (or three) short flashes of each image before advancing to the next frame while the screen was dark.
This is an important concept: Time Shaper can fix camera aliasing, but does nothing about the display or projector. Therefore, it is expected that display "strobing" may still be present even in the smoothest motion settings where camera "judder" has been eliminated completely.
60 Frame Per Second Output
The 60 frame per second demonstration clip is a 970 MB h.264 file. There is no Vimeo embed because Vimeo does not support 60 fps yet.
One way of reducing "strobing" due to the display device is to use a higher delivery frame rate to the projector. In this demonstration, the same 120 fps sequence is used to create 60 fps outputs, and some outputs with a 24fps "look".
Square Wave 180º Shutter
As a baseline, we start with a standard 180º square shutter. This is achieved by simply using every other frame from the 120 fps imagery, and is an option in Time Shaper. Similarly, two frames could be averaged together to achieve a 360º square shutter (commonly used in television), but this is not shown here.
One thing to observe with the 180º shutter is the amount of judder present. Especially in the human motion test, if you keep you eye on the subject as she jogs past the background, you will see enormous amounts of judder present in the background. This simple test shows that increasing frame rate does nothing to reduce judder; judder is entirely the result of the shape of the shutter regardless of frame rate.
Tessive Magnesium and Beryllium Resample
The Magnesium and Beryllium resampling tests provide crisp motion and reduces aliasing while retaining the "60 fps" look. They have response similar to a 180º square shutter but with reduced judder. Because it uses the entire frequency range of 60 fps, they do very little to reduce projector or display strobing (which can be handled separately by the display, but usually isn't.)
24 fps Looks at 60 fps
This demonstration has three looks that demonstrate using 60 fps delivery of 24 fps bandwidth. From the 120 fps input, motion is bandwidth-limited to the response of 24fps, but this is computed for every output frame at 60 fps. At no point is the data actually reduced to 24 frames per second, but rather the frequency response is limited to match that of 24 frames per second. The advantage of doing this is that the 24 fps "look" can be retained, but both camera judder and display strobing can both be nearly eliminated.
The first technique is a 540º square shutter. Time Shaper offers synthetic shutters wider than 360º by allowing overlapping averaging of the incoming frames. At 60 frames per second, a 540º shutter is a 1/40th of a second square wave, very close in exposure time to the traditional 1/48th second 24fps 180º shutter. Because it is a square shutter, it retains the judder of 24fps, but since it is computed exactly for each output frame, there is very little display strobing.
The graph shows the frequency response curves for the various techniques as well as the Nyquist limits for 24 and 60 fps. If we desire a 24 fps look, we want to match the frequency response, but we'd like to eliminate any frequency response above the Nyquist rate. Notice that a 24 fps 180º shutter (red line) has substantial contrast response above the 24 fps Nyquist rate (dotted line at 12 Hz.) This undesirable response is what causes judder. When creating 60 fps output, we have the flexibility to roll off the contrast response in the region between 12 Hz and the 30 Hz new Nyquist rate. The green line shows the Tessive Magnesium response curve, which follows quite closely the 24 fps 180º line but has much less aliasing. The blue line is the Tessive Gallium shutter response, and it has most of the contrast of the 24 fps curve up to 12 Hz, but almost no aliasing past 30 Hz.
Time Shaper can display these curves to aid in comparison of different looks. When evaluating different looks, it is often nice to refer to these curves to help understand what frequency ranges map to various perceived differences.