SLIDE 1
. . . . . .
. . . . . .
Animation Sequence Compression Yang Liu Department of Computer - - PowerPoint PPT Presentation
Animation Sequence Compression Yang Liu Department of Computer - - PowerPoint PPT Presentation
Animation Sequence Compression Yang Liu Department of Computer Science March 2009 . . . . . . Outline Introduction Purpose Animation Sequence Problem Definition Theory about Compression Techniques Techniques in Animation Compression
SLIDE 2
SLIDE 3
. . . . . .
Outline
Introduction Purpose Animation Sequence Problem Definition Theory about Compression Techniques Techniques in Animation Compression
SLIDE 4
. . . . . .
Purpose
◮ Introduction of Animation Sequence Compression
SLIDE 5
. . . . . .
Purpose
◮ Introduction of Animation Sequence Compression ◮ Introduction of Related Techniques
SLIDE 6
. . . . . .
Outline
Introduction Purpose Animation Sequence Problem Definition Theory about Compression Techniques Techniques in Animation Compression
SLIDE 7
. . . . . .
Sample Application
◮ Interactive Entertainment
SLIDE 8
. . . . . .
Sample Application
◮ Interactive Entertainment ◮ Games: World of WarCraft
SLIDE 9
. . . . . .
Sample Application
◮ Interactive Entertainment ◮ Games: World of WarCraft ◮ 3D Movies
SLIDE 10
. . . . . .
Sample Application
◮ Interactive Entertainment ◮ Games: World of WarCraft ◮ 3D Movies ◮ and more...
SLIDE 11
. . . . . .
Outline
Introduction Purpose Animation Sequence Problem Definition Theory about Compression Techniques Techniques in Animation Compression
SLIDE 12
. . . . . .
Audio
◮ Audio is sequence of Magnitude.
SLIDE 13
. . . . . .
Audio
◮ Audio is sequence of Magnitude. ◮ 44.1 kHz, etc. (NyquistShannon sampling theorem)
SLIDE 14
. . . . . .
Audio
◮ Audio is sequence of Magnitude. ◮ 44.1 kHz, etc. (NyquistShannon sampling theorem) ◮ Or, sequence of numbers: WAV format
SLIDE 15
. . . . . .
Audio
◮ Audio is sequence of Magnitude. ◮ 44.1 kHz, etc. (NyquistShannon sampling theorem) ◮ Or, sequence of numbers: WAV format ◮ There is lossy compression: MP3
SLIDE 16
. . . . . .
Audio
◮ Audio is sequence of Magnitude. ◮ 44.1 kHz, etc. (NyquistShannon sampling theorem) ◮ Or, sequence of numbers: WAV format ◮ There is lossy compression: MP3 ◮ There is also no-loss compression: FLAC etc.
SLIDE 17
. . . . . .
Image/Video
◮ Image/Video is a similar scenario
SLIDE 18
. . . . . .
Image/Video
◮ Image/Video is a similar scenario ◮ Image is static
SLIDE 19
. . . . . .
Image/Video
◮ Image/Video is a similar scenario ◮ Image is static ◮ Video is sequence of images
SLIDE 20
. . . . . .
Image/Video
◮ Image/Video is a similar scenario ◮ Image is static ◮ Video is sequence of images ◮ Each image is called “frame”. NTSC video contains 30 frames
per second.
SLIDE 21
. . . . . .
Image/Video
◮ Image/Video is a similar scenario ◮ Image is static ◮ Video is sequence of images ◮ Each image is called “frame”. NTSC video contains 30 frames
per second.
◮ MPEG is the most popular video compression method
nowadays.
SLIDE 22
. . . . . .
Image/Video
◮ Image/Video is a similar scenario ◮ Image is static ◮ Video is sequence of images ◮ Each image is called “frame”. NTSC video contains 30 frames
per second.
◮ MPEG is the most popular video compression method
nowadays.
◮ MPEG-1 for VCD, MPEG-2 for DVD, MPEG-4/H.264 for
network video.
SLIDE 23
. . . . . .
Image/Video
◮ Image/Video is a similar scenario ◮ Image is static ◮ Video is sequence of images ◮ Each image is called “frame”. NTSC video contains 30 frames
per second.
◮ MPEG is the most popular video compression method
nowadays.
◮ MPEG-1 for VCD, MPEG-2 for DVD, MPEG-4/H.264 for
network video.
◮ Most of these compression method are lossy.
SLIDE 24
. . . . . .
3D Animation
◮ 3D Models are static
SLIDE 25
. . . . . .
3D Animation
◮ 3D Models are static ◮ 3D Animation is sequence of 3D Models (frames)
SLIDE 26
. . . . . .
3D Animation
◮ 3D Models are static ◮ 3D Animation is sequence of 3D Models (frames) ◮ Each frame may contains more than one 3D objects.
SLIDE 27
. . . . . .
3D Animation
◮ 3D Models are static ◮ 3D Animation is sequence of 3D Models (frames) ◮ Each frame may contains more than one 3D objects. ◮ 3D Objects in adjacent frame are similar, in most case, have
same connectivity.
SLIDE 28
. . . . . .
3D Animation
◮ 3D Models are static ◮ 3D Animation is sequence of 3D Models (frames) ◮ Each frame may contains more than one 3D objects. ◮ 3D Objects in adjacent frame are similar, in most case, have
same connectivity.
◮ 3D Objects may translate, rotate, and deform.
SLIDE 29
. . . . . .
3D Animation
◮ 3D Models are static ◮ 3D Animation is sequence of 3D Models (frames) ◮ Each frame may contains more than one 3D objects. ◮ 3D Objects in adjacent frame are similar, in most case, have
same connectivity.
◮ 3D Objects may translate, rotate, and deform. ◮ Raw data could be extremly large.
SLIDE 30
. . . . . .
3D Animation
◮ 3D Models are static ◮ 3D Animation is sequence of 3D Models (frames) ◮ Each frame may contains more than one 3D objects. ◮ 3D Objects in adjacent frame are similar, in most case, have
same connectivity.
◮ 3D Objects may translate, rotate, and deform. ◮ Raw data could be extremly large. ◮ Compress data for transferring and storage
SLIDE 31
. . . . . .
3D Animation
◮ 3D Models are static ◮ 3D Animation is sequence of 3D Models (frames) ◮ Each frame may contains more than one 3D objects. ◮ 3D Objects in adjacent frame are similar, in most case, have
same connectivity.
◮ 3D Objects may translate, rotate, and deform. ◮ Raw data could be extremly large. ◮ Compress data for transferring and storage ◮ De-compress/Reconstruct data for rendering
SLIDE 32
. . . . . .
Outline
Introduction Purpose Animation Sequence Problem Definition Theory about Compression Techniques Techniques in Animation Compression
SLIDE 33
. . . . . .
Why data is compressible?
◮ Data Compression is based on Information Theory and Coding
Theory
SLIDE 34
. . . . . .
Why data is compressible?
◮ Data Compression is based on Information Theory and Coding
Theory
◮ Information is the measure of entropy for information source
(Information Theory by Shanon)
SLIDE 35
. . . . . .
Why data is compressible?
◮ Data Compression is based on Information Theory and Coding
Theory
◮ Information is the measure of entropy for information source
(Information Theory by Shanon)
◮ H(X) = − ∑ p(x) log p(x)
SLIDE 36
. . . . . .
Why data is compressible?
◮ Data Compression is based on Information Theory and Coding
Theory
◮ Information is the measure of entropy for information source
(Information Theory by Shanon)
◮ H(X) = − ∑ p(x) log p(x) ◮ It is based on probability theory and statistics.
SLIDE 37
. . . . . .
Why data is compressible?
◮ Data Compression is based on Information Theory and Coding
Theory
◮ Information is the measure of entropy for information source
(Information Theory by Shanon)
◮ H(X) = − ∑ p(x) log p(x) ◮ It is based on probability theory and statistics. ◮ This indicates the minimum number of bits required
(maximum compression possible)
SLIDE 38
. . . . . .
Why data is compressible?-cnt
◮ Entropy Coding (Huffman Coding) is a good way to compress.
SLIDE 39
. . . . . .
Why data is compressible?-cnt
◮ Entropy Coding (Huffman Coding) is a good way to compress. ◮ Basic idea is to represent frequently appearing symbols with
less bits.
SLIDE 40
. . . . . .
Why data is compressible?-cnt
◮ Entropy Coding (Huffman Coding) is a good way to compress. ◮ Basic idea is to represent frequently appearing symbols with
less bits.
◮ Arithmetic Coding is the the closest coding method to
- ptimal encoding.
SLIDE 41
. . . . . .
Why data is compressible?-cnt
◮ Entropy Coding (Huffman Coding) is a good way to compress. ◮ Basic idea is to represent frequently appearing symbols with
less bits.
◮ Arithmetic Coding is the the closest coding method to
- ptimal encoding.
◮ Complexity is the problem.
SLIDE 42
. . . . . .
Why data is compressible?-cnt 2
◮ Previously we talked about “independent” variables.
SLIDE 43
. . . . . .
Why data is compressible?-cnt 2
◮ Previously we talked about “independent” variables. ◮ When next symbol is related to previous symbol, we have less
information.
SLIDE 44
. . . . . .
Why data is compressible?-cnt 2
◮ Previously we talked about “independent” variables. ◮ When next symbol is related to previous symbol, we have less
information.
◮ Possible to compress more.
SLIDE 45
. . . . . .
Why data is compressible?-cnt 2
◮ Previously we talked about “independent” variables. ◮ When next symbol is related to previous symbol, we have less
information.
◮ Possible to compress more. ◮ Hidden Markov Model in Audio Compression.
SLIDE 46
. . . . . .
Lossy Compression
◮ We do need exact information, in some cases.
SLIDE 47
. . . . . .
Lossy Compression
◮ We do need exact information, in some cases. ◮ But not always.
SLIDE 48
. . . . . .
Lossy Compression
◮ We do need exact information, in some cases. ◮ But not always. ◮ MP3, JPEG, MPEG, H.264, ... , all lossy
SLIDE 49
. . . . . .
Lossy Compression
◮ We do need exact information, in some cases. ◮ But not always. ◮ MP3, JPEG, MPEG, H.264, ... , all lossy ◮ For CAD/CAE/documents, precision does matter
SLIDE 50
. . . . . .
Lossy Compression
◮ We do need exact information, in some cases. ◮ But not always. ◮ MP3, JPEG, MPEG, H.264, ... , all lossy ◮ For CAD/CAE/documents, precision does matter ◮ For entertainment, not really
SLIDE 51
. . . . . .
Lossy Compression
◮ We do need exact information, in some cases. ◮ But not always. ◮ MP3, JPEG, MPEG, H.264, ... , all lossy ◮ For CAD/CAE/documents, precision does matter ◮ For entertainment, not really ◮ Just drop not-so-important info
SLIDE 52
. . . . . .
Lossy Compression
◮ We do need exact information, in some cases. ◮ But not always. ◮ MP3, JPEG, MPEG, H.264, ... , all lossy ◮ For CAD/CAE/documents, precision does matter ◮ For entertainment, not really ◮ Just drop not-so-important info ◮ Usually done by quantization – Covered later
SLIDE 53
. . . . . .
Outline
Introduction Purpose Animation Sequence Problem Definition Theory about Compression Techniques Techniques in Animation Compression
SLIDE 54
. . . . . .
Techniques in Audio Compression
◮ Basically, audio compression relies on predictor.
SLIDE 55
. . . . . .
Techniques in Audio Compression
◮ Basically, audio compression relies on predictor. ◮ Idea: any magnitude is related to previous magnitude.
SLIDE 56
. . . . . .
Techniques in Audio Compression
◮ Basically, audio compression relies on predictor. ◮ Idea: any magnitude is related to previous magnitude. ◮ Identical predictors work on both ends: Compression and
De-compression.
SLIDE 57
. . . . . .
Techniques in Audio Compression
◮ Basically, audio compression relies on predictor. ◮ Idea: any magnitude is related to previous magnitude. ◮ Identical predictors work on both ends: Compression and
De-compression.
◮ Only residue is stored/transferred.
SLIDE 58
. . . . . .
Techniques in Audio Compression
◮ Basically, audio compression relies on predictor. ◮ Idea: any magnitude is related to previous magnitude. ◮ Identical predictors work on both ends: Compression and
De-compression.
◮ Only residue is stored/transferred. ◮ Residue should be very small.
SLIDE 59
. . . . . .
Techniques in Audio Compression
◮ Basically, audio compression relies on predictor. ◮ Idea: any magnitude is related to previous magnitude. ◮ Identical predictors work on both ends: Compression and
De-compression.
◮ Only residue is stored/transferred. ◮ Residue should be very small. ◮ Entropy coding could compress residue easily
SLIDE 60
. . . . . .
Techniques in Audio Compression-cnt
◮ Predictors for non-lossy compression: FLAC etc.
SLIDE 61
. . . . . .
Techniques in Audio Compression-cnt
◮ Predictors for non-lossy compression: FLAC etc. ◮ MP3: Just drop high-frequency signals.
SLIDE 62
. . . . . .
Techniques in Audio Compression-cnt
◮ Predictors for non-lossy compression: FLAC etc. ◮ MP3: Just drop high-frequency signals. ◮ People are not picky on high-frequency sound
SLIDE 63
. . . . . .
Techniques in Audio Compression-cnt
◮ Predictors for non-lossy compression: FLAC etc. ◮ MP3: Just drop high-frequency signals. ◮ People are not picky on high-frequency sound ◮ Fourier Transform
SLIDE 64
. . . . . .
Techniques in JPEG
◮ Employes Discrete Cosine Transform (Specialized Fourier
Transform).
SLIDE 65
. . . . . .
Techniques in JPEG
◮ Employes Discrete Cosine Transform (Specialized Fourier
Transform).
◮ In JPEG2000, Wavelet Transform is employed.
SLIDE 66
. . . . . .
Techniques in JPEG
◮ Employes Discrete Cosine Transform (Specialized Fourier
Transform).
◮ In JPEG2000, Wavelet Transform is employed. ◮ Drop High-Frequency signal (Details).
SLIDE 67
. . . . . .
Techniques in JPEG
◮ Employes Discrete Cosine Transform (Specialized Fourier
Transform).
◮ In JPEG2000, Wavelet Transform is employed. ◮ Drop High-Frequency signal (Details). ◮ Run-Length Encoding
SLIDE 68
. . . . . .
Techniques in JPEG
◮ Employes Discrete Cosine Transform (Specialized Fourier
Transform).
◮ In JPEG2000, Wavelet Transform is employed. ◮ Drop High-Frequency signal (Details). ◮ Run-Length Encoding ◮ Huffman Coding
SLIDE 69
. . . . . .
Techniques in JPEG
◮ Employes Discrete Cosine Transform (Specialized Fourier
Transform).
◮ In JPEG2000, Wavelet Transform is employed. ◮ Drop High-Frequency signal (Details). ◮ Run-Length Encoding ◮ Huffman Coding ◮ Lossy
SLIDE 70
. . . . . .
Techniques in JPEG
◮ Employes Discrete Cosine Transform (Specialized Fourier
Transform).
◮ In JPEG2000, Wavelet Transform is employed. ◮ Drop High-Frequency signal (Details). ◮ Run-Length Encoding ◮ Huffman Coding ◮ Lossy ◮ Image may be blurred, with high compression ratio.
SLIDE 71
. . . . . .
Techniques in MPEG
◮ Intuitive way to store/transfer video is to transfer all images
- ne by one (Raw Data).
SLIDE 72
. . . . . .
Techniques in MPEG
◮ Intuitive way to store/transfer video is to transfer all images
- ne by one (Raw Data).
◮ By using similarity between adjacent image, it is possible to
remove redundant information, thus compress data.
SLIDE 73
. . . . . .
Techniques in MPEG
◮ Intuitive way to store/transfer video is to transfer all images
- ne by one (Raw Data).
◮ By using similarity between adjacent image, it is possible to
remove redundant information, thus compress data.
◮ Another way is to “drop” some not so important information.
That is, lossy compression.
SLIDE 74
. . . . . .
Techniques in MPEG
◮ Intuitive way to store/transfer video is to transfer all images
- ne by one (Raw Data).
◮ By using similarity between adjacent image, it is possible to
remove redundant information, thus compress data.
◮ Another way is to “drop” some not so important information.
That is, lossy compression.
◮ MPEG is based on JPEG. Both of them are lossy to gain more
compression ratio.
SLIDE 75
. . . . . .
Techniques in MPEG - cnt
◮ Normally, video contains 24 - 30 frams per second
(Film/NTSC)
SLIDE 76
. . . . . .
Techniques in MPEG - cnt
◮ Normally, video contains 24 - 30 frams per second
(Film/NTSC)
◮ In-Frame Compression: JPEG
SLIDE 77
. . . . . .
Techniques in MPEG - cnt
◮ Normally, video contains 24 - 30 frams per second
(Film/NTSC)
◮ In-Frame Compression: JPEG ◮ I-frame and P-frame
SLIDE 78
. . . . . .
Techniques in MPEG - cnt
◮ Normally, video contains 24 - 30 frams per second
(Film/NTSC)
◮ In-Frame Compression: JPEG ◮ I-frame and P-frame ◮ I-frame is stored/transferred completely (high-quality)
SLIDE 79
. . . . . .
Techniques in MPEG - cnt
◮ Normally, video contains 24 - 30 frams per second
(Film/NTSC)
◮ In-Frame Compression: JPEG ◮ I-frame and P-frame ◮ I-frame is stored/transferred completely (high-quality) ◮ P-frame is stroed/transferred as residue (low-quality)
SLIDE 80
. . . . . .
Techniques in MPEG - cnt-2
◮ What the coder do:
SLIDE 81
. . . . . .
Techniques in MPEG - cnt-2
◮ What the coder do: ◮ Find blocks moving on screen and background
SLIDE 82
. . . . . .
Techniques in MPEG - cnt-2
◮ What the coder do: ◮ Find blocks moving on screen and background ◮ Compress background and moving objects seperately
SLIDE 83
. . . . . .
Techniques in MPEG - cnt-2
◮ What the coder do: ◮ Find blocks moving on screen and background ◮ Compress background and moving objects seperately ◮ For I-frame, just store/transfer it.
SLIDE 84
. . . . . .
Techniques in MPEG - cnt-2
◮ What the coder do: ◮ Find blocks moving on screen and background ◮ Compress background and moving objects seperately ◮ For I-frame, just store/transfer it. ◮ For P-frame, predicts movement of blocks, store/transfer
residue
SLIDE 85
. . . . . .
Techniques in MPEG - cnt-2
◮ What the coder do: ◮ Find blocks moving on screen and background ◮ Compress background and moving objects seperately ◮ For I-frame, just store/transfer it. ◮ For P-frame, predicts movement of blocks, store/transfer
residue
◮ For blocks, it may change. Store/transfer the residue of block
SLIDE 86
. . . . . .
Techniques in MPEG - cnt-3
◮ Finding moving blocks is hard. So encoding takes more time
than decoding.
SLIDE 87
. . . . . .
Techniques in MPEG - cnt-3
◮ Finding moving blocks is hard. So encoding takes more time
than decoding.
◮ That’s why different MPEG encoder may produce different
compression result
SLIDE 88
. . . . . .
Techniques in MPEG - cnt-3
◮ Finding moving blocks is hard. So encoding takes more time
than decoding.
◮ That’s why different MPEG encoder may produce different
compression result
◮ More residue information means better quality and more bits
SLIDE 89
. . . . . .
Techniques in MPEG - cnt-3
◮ Finding moving blocks is hard. So encoding takes more time
than decoding.
◮ That’s why different MPEG encoder may produce different
compression result
◮ More residue information means better quality and more bits ◮ Less residue information means lower quality and less bits
SLIDE 90
. . . . . .
Techniques in MPEG - cnt-3
◮ Finding moving blocks is hard. So encoding takes more time
than decoding.
◮ That’s why different MPEG encoder may produce different
compression result
◮ More residue information means better quality and more bits ◮ Less residue information means lower quality and less bits ◮ That’s why action movie requires more bits.
SLIDE 91
. . . . . .
Outline
Introduction Purpose Animation Sequence Problem Definition Theory about Compression Techniques Techniques in Animation Compression
SLIDE 92
. . . . . .
Animation Compression
◮ Predictor-Based method:
SLIDE 93
. . . . . .
Animation Compression
◮ Predictor-Based method: ◮ In-frame compression: Space-only predictor
SLIDE 94
. . . . . .
Animation Compression
◮ Predictor-Based method: ◮ In-frame compression: Space-only predictor ◮ That is, transfer only part of vertices.
SLIDE 95
. . . . . .
Animation Compression
◮ Predictor-Based method: ◮ In-frame compression: Space-only predictor ◮ That is, transfer only part of vertices. ◮ Inter-frame compression: Time-only predictor and residue for
each vertex
SLIDE 96
. . . . . .
Animation Compression
◮ Predictor-Based method: ◮ In-frame compression: Space-only predictor ◮ That is, transfer only part of vertices. ◮ Inter-frame compression: Time-only predictor and residue for
each vertex
◮ Combine them together: Space-Time predictor
SLIDE 97
. . . . . .
Animation Compression
◮ Predictor-Based method: ◮ In-frame compression: Space-only predictor ◮ That is, transfer only part of vertices. ◮ Inter-frame compression: Time-only predictor and residue for
each vertex
◮ Combine them together: Space-Time predictor ◮ Using quantization, we can compress model with degraded
quality.
SLIDE 98
. . . . . .
Animation Compression
◮ Predictor-Based method: ◮ In-frame compression: Space-only predictor ◮ That is, transfer only part of vertices. ◮ Inter-frame compression: Time-only predictor and residue for
each vertex
◮ Combine them together: Space-Time predictor ◮ Using quantization, we can compress model with degraded
quality.
◮ Assumption: Connectivity never change.
SLIDE 99
. . . . . .
Animation Compression
◮ Predictor-Based method: ◮ In-frame compression: Space-only predictor ◮ That is, transfer only part of vertices. ◮ Inter-frame compression: Time-only predictor and residue for
each vertex
◮ Combine them together: Space-Time predictor ◮ Using quantization, we can compress model with degraded
quality.
◮ Assumption: Connectivity never change. ◮ Dynapack: Space-Time compression of the 3D animations of
triangle meshes with fixed connectivity, Lawrence Ibarria et al.
SLIDE 100
. . . . . .
Animation Compression-cnt
◮ Skeleton-Based method:
SLIDE 101
. . . . . .
Animation Compression-cnt
◮ Skeleton-Based method: ◮ For many models, such as human, there are skeletons.
SLIDE 102
. . . . . .
Animation Compression-cnt
◮ Skeleton-Based method: ◮ For many models, such as human, there are skeletons. ◮ Skeleton has much less degree of freedom (DOF).
SLIDE 103
. . . . . .
Animation Compression-cnt
◮ Skeleton-Based method: ◮ For many models, such as human, there are skeletons. ◮ Skeleton has much less degree of freedom (DOF). ◮ Normally skeletons just rotate and translate.
SLIDE 104
. . . . . .
Animation Compression-cnt
◮ Skeleton-Based method: ◮ For many models, such as human, there are skeletons. ◮ Skeleton has much less degree of freedom (DOF). ◮ Normally skeletons just rotate and translate. ◮ Thus we can use sequence of rotation quaternions and
translation vectors to represent each frame.
SLIDE 105
. . . . . .
Animation Compression-cnt
◮ Skeleton-Based method: ◮ For many models, such as human, there are skeletons. ◮ Skeleton has much less degree of freedom (DOF). ◮ Normally skeletons just rotate and translate. ◮ Thus we can use sequence of rotation quaternions and
translation vectors to represent each frame.
◮ Then it comes to compression of quaternions and vectors,
which is much easier.
SLIDE 106
. . . . . .
Animation Compression-cnt
◮ Skeleton-Based method: ◮ For many models, such as human, there are skeletons. ◮ Skeleton has much less degree of freedom (DOF). ◮ Normally skeletons just rotate and translate. ◮ Thus we can use sequence of rotation quaternions and
translation vectors to represent each frame.
◮ Then it comes to compression of quaternions and vectors,
which is much easier.
◮ That’s why in most game animation has no deformation.
SLIDE 107
. . . . . .
Animation Compression-cnt
◮ Skeleton-Based method: ◮ For many models, such as human, there are skeletons. ◮ Skeleton has much less degree of freedom (DOF). ◮ Normally skeletons just rotate and translate. ◮ Thus we can use sequence of rotation quaternions and
translation vectors to represent each frame.
◮ Then it comes to compression of quaternions and vectors,
which is much easier.
◮ That’s why in most game animation has no deformation. ◮ It’s possible to apply Fourier Transform on quaternions.
SLIDE 108
. . . . . .
Animation Compression-cnt
◮ Skeleton-Based method: ◮ For many models, such as human, there are skeletons. ◮ Skeleton has much less degree of freedom (DOF). ◮ Normally skeletons just rotate and translate. ◮ Thus we can use sequence of rotation quaternions and
translation vectors to represent each frame.
◮ Then it comes to compression of quaternions and vectors,
which is much easier.
◮ That’s why in most game animation has no deformation. ◮ It’s possible to apply Fourier Transform on quaternions. ◮ Efficient Implementation of Quaternion Fourier Transform,
Convolution, and Correlation by 2-D Complex FFT, Soo-Chang Pei et al.
SLIDE 109
. . . . . .
Animation Compression-cnt 2
◮ What if there is no skeleton?
SLIDE 110
. . . . . .
Animation Compression-cnt 2
◮ What if there is no skeleton? ◮ Find relative patches by statistics
SLIDE 111
. . . . . .
Animation Compression-cnt 2
◮ What if there is no skeleton? ◮ Find relative patches by statistics ◮ Then use these patch to represent animations
SLIDE 112
. . . . . .
Animation Compression-cnt 2
◮ What if there is no skeleton? ◮ Find relative patches by statistics ◮ Then use these patch to represent animations ◮ Add more details progressively.
SLIDE 113
. . . . . .
Animation Compression-cnt 2
◮ What if there is no skeleton? ◮ Find relative patches by statistics ◮ Then use these patch to represent animations ◮ Add more details progressively. ◮ Skinning Mesh Animation, Doug L. James
SLIDE 114
. . . . . .
Animation Compression-cnt 3
◮ Decomposition-based compression:
SLIDE 115
. . . . . .
Animation Compression-cnt 3
◮ Decomposition-based compression: ◮ Decompose the difference between “reference” frame and
“other frame”
SLIDE 116
. . . . . .
Animation Compression-cnt 3
◮ Decomposition-based compression: ◮ Decompose the difference between “reference” frame and
“other frame”
◮ Add difference progressively
SLIDE 117
. . . . . .
Animation Compression-cnt 3
◮ Decomposition-based compression: ◮ Decompose the difference between “reference” frame and
“other frame”
◮ Add difference progressively ◮ Use quantization to compromise between bits and quality
SLIDE 118
. . . . . .
Animation Compression-cnt 4
◮ Geometry Video:
SLIDE 119
. . . . . .
Animation Compression-cnt 4
◮ Geometry Video: ◮ It is based on Geometry Image: Convert 3D models to 2D
image by generating regular connectivity.
SLIDE 120
. . . . . .
Animation Compression-cnt 4
◮ Geometry Video: ◮ It is based on Geometry Image: Convert 3D models to 2D
image by generating regular connectivity.
◮ Putting all images for frames together, we have a video.
SLIDE 121
. . . . . .
Animation Compression-cnt 4
◮ Geometry Video: ◮ It is based on Geometry Image: Convert 3D models to 2D
image by generating regular connectivity.
◮ Putting all images for frames together, we have a video. ◮ Compress it as video.
SLIDE 122
. . . . . .
Animation Compression-cnt 4
◮ Geometry Video: ◮ It is based on Geometry Image: Convert 3D models to 2D
image by generating regular connectivity.
◮ Putting all images for frames together, we have a video. ◮ Compress it as video. ◮ Geometry videos: a new representation for 3D animations,
Hector M Briceno et al.
SLIDE 123
. . . . . .
Thank you
◮ More details about techniques will be covered next time.
SLIDE 124
. . . . . .