Synthetic Examples: Astronaut Sequence with Rotation

Back to the index

Video

The astronaut sequence with rotation, played back at 10 frames per second. The sequence has a variable exposure time with noise added to simulate adjustments to the gain. Click on the video to start playing.

Sequence spline

The piecewise cubic spline used to generate the sequence. The foreground moves along the spline from right to left while rotating about its center. Heavier segments of the curve indicate the exposures used to simulate motion blur.

Pairwise flows

Frame 6 to frame 5

In this example, the source (frame 5) is sharp, and the target (frame 6) is blurred.

Please move your mouse cursor over the small images to toggle between the results.

Target Source Baseline Our method Ground truth

Frame 6 (target)

Frame 5 (source)

Baseline flow

Our flow

Ground truth flow

Frame 6 (target)

Frame 6 (target)

Our flow is much more consisent and better resembles the ground-truth flow than that produced by the baseline method.

Warping results from the source image (frame 5) to the target image (frame 6) using estimated flows are shown below.

Source Baseline Our method Target

Input source (frame 5)

Warped from input, baseline

Warped from input, our method

Input target (frame 6)

True target (frame 6)

Ground-truth
target
Input source (frame 5)

Input source (frame 5)

Our method better preserves the sharpness and structure of the source image.

Frame 3 to frame 2

In this example, both source (frame 2) and target (frame 3) are blurred.

Target Source Baseline Our method Ground truth

Frame 3 (target)

Frame 2 (source)

Baseline flow

Our flow

Ground truth flow

Frame 3 (target)

Frame 3 (target)

The baseline flow appears to be very inconsistent compared to our flow. Both flows are inaccurate in the upper-right corner where the background has little texture to track so the smoothness term dominates.

Warping results from the source image (frame 2) to the target image (frame 3) using estimated flows are shown below.

Source Baseline Our method Target
Warping using
input frames

Input source (frame 2)

Warped from input, baseline

Warped from input, our method

Input target (frame 3)

Warping using
“ground-truth”
frames

True source (frame 2)

Warped from truth, baseline

Warped from truth, our method

True target (frame 3)

Input source (frame 2)

Input source (frame 2)

The baseline method tries to warp the source to match the blur direction of the target, whereas our method preserves the appearance of the source. Our result is preferable, as indicated by a more accurate warping using the ground truth source and target.

Concatenated flow

Frame 7 to frame 11

In this example, both the source (frame 11) and the target (frame 7) are sharp. They are separated by multiple blurred frames.

Source Baseline Our method Target

Source (frame 11)

Warped, baseline

Warped, our method

Target (frame 7)

Source (frame 11)

Source (frame 11)

Our flow is more accurate, as demonstrated by the warping result more closely matching the target.

Back to the index