Decomposing Time-Lapse Paintings into Layers

This code implements the pipeline described in the paper "Decomposing Time-Lapse Paintings into Layers" by Jianchao Tan, Marek Dvorožňák, Daniel Sýkora, Yotam Gingold from SIGGRAPH 2015. The pipeline is divided into two stages.

1 Preprocessing:

• Input: raw time-lapse video
• Output: albedo video

The substeps are:

1. Color shift the whole sequence
2. Extract keyframes and color shift each sub-sequence
3. For each sub-sequence, perform moving std. deviation and moving median
4. Whole sequence L0 smoothing
5. Perform albedo conversion

2 Layer extraction

• Input: albedo video
• Output: KM layers and PD layers

The programs are:

• PD layer extraction and KM layer extraction
• PD using the spatial coherency solution: The 3-by-3 layer extraction described in the paper

Dependencies

• OpenCV 2.4
• Eigen 3
• JsonCpp 0.5
• zlib
• Bottleneck: pip install bottleneck
• PIL or Pillow (Python Image Library): pip install Pillow
• NumPy
• LAPACK