Index of /~gwtaylor/publications/nips2006mhmublv/videos
Name Last modified Size Description
![[DIR]](/icons/back.gif)
Parent Directory -
![[DIR]](/icons/folder.gif)
geoff/ 20-Oct-2008 16:06 -
![[Video]](/icons/movie.gif)
1walking.gif 26-Sep-2006 14:05 556K Video: generated motion
1walking.mp4 25-Sep-2008 13:08 888K
2running.gif 26-Sep-2006 14:05 214K Video: generated motion
2running.mp4 25-Sep-2008 13:08 397K
3run_to_walk_noise.gif 26-Sep-2006 14:05 430K Video: generated motion
3run_to_walk_noise.mp4 25-Sep-2008 13:08 863K
4run_to_walk.gif 26-Sep-2006 14:05 2.0M Video: generated motion
4run_to_walk.mp4 25-Sep-2008 13:08 2.1M
5longwalk.gif 26-Sep-2006 14:05 1.4M Video: generated motion
5longwalk.mp4 25-Sep-2008 13:08 3.5M
6walk_pause_walk.gif 26-Sep-2006 14:05 1.4M Video: generated motion
6walk_pause_walk.mp4 25-Sep-2008 13:08 2.0M
7filling.gif 26-Sep-2006 14:05 2.6M Video: generated motion
7filling.mp4 25-Sep-2008 13:08 539K
A_walk.gif 18-Dec-2006 17:19 672K Video: generated motion
A_walk.mp4 25-Sep-2008 13:08 1.2M
B_run.gif 18-Dec-2006 17:19 1.1M Video: generated motion
B_run.mp4 25-Sep-2008 13:08 1.7M
C_transition.gif 18-Dec-2006 17:19 1.0M Video: generated motion
C_transition.mp4 25-Sep-2008 13:08 1.4M
D_120.gif 18-Dec-2006 17:20 2.7M Video: generated motion
D_120.mp4 25-Sep-2008 13:08 1.1M
cat.gif 24-Sep-2008 17:14 4.2M Video: generated motion
cat.mp4 24-Sep-2008 17:14 1.2M
![[IMG]](/icons/image2.gif)
cat.png 24-Sep-2008 17:20 9.4K
![[TXT]](/icons/text.gif)
cat.xml 25-Sep-2008 12:50 672
chicken.gif 24-Sep-2008 17:14 4.9M Video: generated motion
chicken.mp4 24-Sep-2008 17:14 1.7M
chicken.png 24-Sep-2008 17:20 12K
chicken.xml 25-Sep-2008 12:50 676
dinosaur.gif 24-Sep-2008 17:14 5.2M Video: generated motion
dinosaur.mp4 24-Sep-2008 17:14 1.7M
dinosaur.png 24-Sep-2008 17:20 13K
dinosaur.xml 25-Sep-2008 12:50 677
drunk.gif 24-Sep-2008 17:14 5.3M Video: generated motion
drunk.mp4 24-Sep-2008 17:14 1.8M
drunk.png 24-Sep-2008 17:20 13K
drunk.xml 25-Sep-2008 12:51 674
gangly.gif 24-Sep-2008 17:15 4.6M Video: generated motion
gangly.mp4 24-Sep-2008 17:15 1.6M
gangly.png 24-Sep-2008 17:20 11K
gangly.xml 25-Sep-2008 12:50 675
graceful.gif 24-Sep-2008 17:15 4.4M Video: generated motion
graceful.mp4 24-Sep-2008 17:15 1.3M
graceful.png 24-Sep-2008 17:20 11K
graceful.xml 25-Sep-2008 12:47 748
normal.gif 24-Sep-2008 17:15 5.3M Video: generated motion
normal.mp4 24-Sep-2008 17:15 1.7M
normal.png 24-Sep-2008 17:20 12K
normal.xml 25-Sep-2008 12:51 670
sexy.gif 24-Sep-2008 17:15 5.1M Video: generated motion
sexy.mp4 24-Sep-2008 17:15 1.6M
sexy.png 24-Sep-2008 17:20 12K
sexy.xml 25-Sep-2008 12:51 668
strong.gif 24-Sep-2008 17:15 4.5M Video: generated motion
strong.mp4 24-Sep-2008 17:15 1.4M
strong.png 24-Sep-2008 17:20 11K
strong.xml 25-Sep-2008 12:51 670
Description of videos:
Videos A-D are generated using the framework discussed in the NIPS paper.
A) The motion was generated using a latent variable model with 200 binary
stochastic units. The model was trained on 3825 frames of motion capture data
consisting of walking sequences. For generation, the model has been initialized
by a few frames of walking.
B) The motion was generated using a latent variable model with 200 binary
stochastic units. The model was trained on 2515 frames of motion capture data
consisting of walking and running sequences. For generation, the model has been
initialized by a few frames of running.
C) The motion was generated using the same model as B), but the model has been
initialized by a few frames of walking. The sequence features a transition from
walking to running.
D) The motion was generated using a latent variable model with 400 binary
stochastic units. The model was trained on 15300 frames of motion capture data
consisting of walking sequences (same as A but not downsampled). This
demonstrates that we are able to learn on 120fps motion (all other sequences are
30fps).
Videos 1-7 are generated using a simplified version of the model discussed in
the NIPS paper. The model has been trained using a 3-stage procedure and
we consider these results to be inferior to the results using our latest
framework.
1) The motion was generated using a latent variable model with 400 binary
stochastic units. The model was trained on 2813 frames of motion capture data
consisting of walking and running sequences. For generation, the model has been
initialized by a few frames of walking.
2) Same model as above, but the model has been initialized by a few frames of
running.
3) Same model as above. The training data contains no transitions between
walking and running or vice-versa, but by inserting a small amount of Gaussian
noise during sampling, the mode occasionally will transition between styles.
4) Here the model has been trained on walking and running sequences that do
contrain transitions between modes. Ocassionally transitions will happen
"naturally" during generated sequences.
5) A long walking sequence generated by a model trained on a different dataset,
(walking sequences only) again with 400 binary stochastic units. During the
generation, we use the real-valued past "probabilities" of the hidden units as
input to the directed connections instead of the stochastically chosen past
activations.
6) The same model as above, but using binary past states instead of real values.
The walking is less smooth, but the behaviour is more "stochastic" and will
occasionally pause, turn around, etc... (there are such pauses in the training
set).
7) The model also has the ability to fill in missing data online. Here, we have
deleted data (left leg, upper body) halfway through the sequence and the model
fills in the missing joint angles.