understanding the semantic part structure of sketches
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
Sandbox
analysis
experiments
results
.gitignore
LICENSE
README.md

README.md

semantic_parts

Our goal is to understand the semantic part structure of object sketches.

Developing part segmentation task on toy case

Milestones for part segmentation task interface:

  1. Display: paper.js canvas environment that displays the sample smiley SVG, by reading in the cubic bezier curve segments.
  2. Interactive 1: allows for highlighting by hovering/clicking particular strokes
  3. Interactive 2: allows you to "paint in" only the subset of curve segments that you click and drag over, rather than entire stroke by default.
  4. Interactive 3: class-conditional menu of labels appears after each "paint" operation (e.g., eye, mouth, head, Other). a. If "Other" is selected, text box appears to allow for free response.
  5. Output: For each paint+annotation cycle, write out: painted SVG curve segments plus their semantic label. Once an SVG curve segment is painted and labeled once, it cannot be painted in again.

Generalizing to a particular real object class

Using part annotations to train a segmenter on held out objects within that class.