ShapeComp

an image-based metric of human shape similarity

Here is our repository for ShapeComp, an image-computable model that is highly predictive of human shape similarity. The repository includes several MATLAB demos that show how to evaluate shapes on ShapeComp and arrange them in terms of similarity. One can also use the repository to create novel shapes, using a GAN trained on >25,000 animal shapes.

What can I do with these tools?

  1. Create novel GAN silhouettes, and compute their shape similarity

  2. Evaluate perceptual shape similarity on your custom shape set

  3. Create novel shape sets (or object classes) controlled for perceptual shape similarity

Requirements

Download:

ShapeComp

Other requirements:

MATLAB

Deep Learning Toolbox

Keras Importer Add on

Demos

For now, we include two demos.

Demo 1

Create novel GAN shapes and then use ShapeComp to arrange them based on their predicted shape similarity.

Demo 2

Evaluate ShapeComp to predict shape similarity on pre-determined shape sets (e.g., shapes from your own experiment). In the example above, we arrange shapes from the Validated Circular Shape Data Set. In other examples, we show how to trace contours from natural images to generate a silhouette that can be fed into ShapeComp.

We plan on including more demos soon (e.g., creating perceptually uniform shape spaces), so please stay tuned....

References

If you use this code, please cite the following:

Morgenstern, Y., Hartmann, F., Schmidt, F., Tiedemann, H., Prokott, E., Maiello, G., & Fleming, R. W. (2021). An image-computable model of human visual shape similarity. PLoS Computational Biology.

Questions

For related work, please visit the sites of Roland Fleming, Filipp Schmidt, and/or Yaniv Morgenstern

For questions or suggestions, please contact Yaniv Morgenstern (Yaniv.Morgenstern@gmail.com).