Single Image Animation: Difference between revisions
| (One intermediate revision by the same user not shown) | |||
| Line 43: | Line 43: | ||
** Residuals are original frame - gaussian blurred frame | ** Residuals are original frame - gaussian blurred frame | ||
===Using Neural Networks=== | |||
* Animating Pictures with Eulerian Motion Fields (2020) | |||
** [https://eulerian.cs.washington.edu/ https://eulerian.cs.washington.edu/] | |||
** Authors: Aleksander Holynski, Brian Curless, Steven M. Seitz, Richard Szeliski | |||
** Warps deep features using a predicted motion field. | |||
** Uses a symmetric splatting and blending technique to fill in holes in the feature maps. | |||
==Animals== | ==Animals== | ||
Latest revision as of 22:14, 3 February 2021
Research on how to create animations from a single image.
Most existing work here is focused on transfering an animation from a video or collection of videos onto an image.
Fluids
Animating Fluids
Relevent People
Animating Pictures of Fluid using Video Examples
- Published in Computer Graphics Forum 2009
- Authors: Makoto Okabe Ken Anjyo Takeo Igarashi Hans‐Peter Seidel
- https://onlinelibrary.wiley.com/doi/full/10.1111/j.1467-8659.2009.01408.x
- https://www.youtube.com/watch?v=RBK6UGz3GPc
- Basic Idea
- Given an image and a video
- Decompose video into average image, flow field, residual
- User draws desired flow field for the image and transfers (?) the residual from the video.
- New animation is average of image + drawn flow field + transfered residual
Creating Fluid Animation from a Single Image using Video Database
- Published in Computer Graphics Forum 2011
- Authors: Makoto Okabe, Ken Anjyo, Rikio Onai
- https://onlinelibrary.wiley.com/doi/full/10.1111/j.1467-8659.2011.02062.x
- https://www.youtube.com/watch?v=S5iqKa7FRy0
- Basic Idea
- Create database of one type of fluid (fire, smoke, water)
- Database is composed of (average image + differences) of video patches and indexed with visual words (SIFT).
- Given a new image of a fluid, cut it up into pieces.
- For each piece, find a video from the database. Combine the differences from the database with the image to form an animation. Lots of small details in the paper...
- User can draw a flow field and apply a mask. The flow field influences which video is selected from the database.
Animating pictures of water scenes using video retrieval
- Published in The Visual Computer 2018
- First online in 2016
- Authors: Makoto Okabe, Yoshinori Dobashi, Ken Anjyo
- https://link.springer.com/content/pdf/10.1007%2Fs00371-016-1337-6.pdf
- https://download-tls-cdn.edge-cdn.net/videodb/5501/videodb_5501_53500_9324610_16x9_hq.mp4
- Basic Idea
- Given an input image
- User draws regions of interests
- For each ROI, find a video with a similar structure to the ROI
- Merge patches and video by coping residuals
- Residuals are original frame - gaussian blurred frame
Using Neural Networks
- Animating Pictures with Eulerian Motion Fields (2020)
- https://eulerian.cs.washington.edu/
- Authors: Aleksander Holynski, Brian Curless, Steven M. Seitz, Richard Szeliski
- Warps deep features using a predicted motion field.
- Uses a symmetric splatting and blending technique to fill in holes in the feature maps.
Animals
Animating Animal Motion from Still
- Siggraph Asia 2008
- Authors: Xuemiao Xu, Liang Wan, Xiaopei Liu, Tien-Tsin Wong, Liansheng Wang, Chi-Sing Leung
- https://www.cse.cuhk.edu.hk/~ttwong/papers/flock/flock.html
- Basic Idea
- Given a photo with multiple of the same animal
- Crop each animal to get a set of "motion snapshots"
- Using some structural similarity measures, order the motion snapshots to create a motion cycle
- Do consistency refinement to morph between snapshots.