1
00:00:01,834 --> 00:00:06,434
Our model for disparity remapping is
built on perceptual experiments.

2
00:00:08,261 --> 00:00:15,561
In those paricipants had to decide
whether stimulus is moving in depth or not.

3
00:00:17,140 --> 00:00:22,073
Or whether its disparity
corrugation is being scaled.

4
00:00:28,130 --> 00:00:32,930
From that we have learnt
threshold speeds for our model.

5
00:00:33,579 --> 00:00:38,446
We use them to power our applications
such as the gaze-driven remapping.

6
00:00:38,807 --> 00:00:43,607
Here you see a static linear mapping of
disparic and gaze as a white dot.

7
00:00:44,098 --> 00:00:48,198
Such mapping does not adapt to
a gaze location change in any way.

8
00:00:48,592 --> 00:00:53,792
An alternative shown here shifts
the disparity at the gaze to the screen.

9
00:00:54,318 --> 00:00:56,918
That, however, leads to annoying jumps.

10
00:01:00,332 --> 00:01:07,465
Hanhart et al. proposed a temporally
coherent extension to improve stability.

11
00:01:12,382 --> 00:01:15,882
Our method takes
scaling of depth into account.

12
00:01:16,264 --> 00:01:19,064
That opens much more
mapping possibilities.

13
00:01:19,135 --> 00:01:23,135
The model ensures that
no temporal artifacts are introduced.

14
00:01:28,081 --> 00:01:32,614
Here we present resulting mappings
as recorded using an eye tracker.

15
00:01:32,792 --> 00:01:36,459
Note that this cannot reproduce
a real user experience.

16
00:01:36,848 --> 00:01:38,915
First we show a static mapping.

17
00:01:41,564 --> 00:01:43,631
Then our gaze adaptive mapping.

18
00:01:44,044 --> 00:01:47,977
You may follow the gaze dot and
note the depth enhancement.

19
00:01:53,624 --> 00:01:57,824
And finally a comparison of
all 4 method using a splitted screen.

20
00:01:58,183 --> 00:02:02,783
Our method does not show
artifacts like the immediate shift.

21
00:02:03,164 --> 00:02:07,097
It also leads to bigger depth impression
than alternatives.

22
00:02:07,536 --> 00:02:10,403
Please refer to the paper for a user study.

23
00:02:10,821 --> 00:02:14,421
Here some more results for both CG and
real content.

24
00:03:54,461 --> 00:03:58,094
Our transition model is also useful
to reduce abrupt disparity changes...

25
00:03:58,210 --> 00:04:01,077
...in movie cuts such as this one.

26
00:04:01,246 --> 00:04:05,446
The disparity change is seamlessly
distributed to other frames.

27
00:04:05,876 --> 00:04:06,876
A comparison.

28
00:04:11,353 --> 00:04:14,286
We also support
an offline depth preprocessing.

29
00:04:15,926 --> 00:04:20,126
Per-frame constructed mapping curves are
temporally incoherent.

30
00:04:27,576 --> 00:04:32,303
Our model finds new curves and
seamless transitions between them.

31
00:04:34,912 --> 00:04:38,122
A comparison.

32
00:04:47,176 --> 00:04:49,676
Thank you.