Heliotrope

UCL Computational Photography and Capture - Heliotrope Lab

Folder structure

• Place optical flow data in data/{dataset_name}_flows.mat
  • For example data/gabe_flows.mat or data/trump_flows.mat
  • You can download the Trump optical flow data from Dropbox (1.83 GB)
• Image sequence should be in the folder data/{dataset_name}
  • E.g. data/gabe/ or data/trump/

Usage

• Run heliotrope
  • Edit the variable dataset in heliotrope.m if you wish to change the dataset
• Enter your desired starting image, or use the default recommended one by just pressing Enter
• Draw as many line segments as you want
• Backspace to delete previous point
• Double-click your final point to end selection
• Wait a minute...

TIP: Use the tip of the nose as starting point to deliver best results.

Output

3 video figures will appear:

• Slow motion interpolation of trajectory-based cost
• No interpolation - Without trajectory cost
• No interpolation - With trajectory cost

Adjustable parameters

• Inside src/best_path.m you can edit alpha to modify the weighting of the trajectory cost.

Generate optical flow

• Run generate_flow