Code

You're looking at the code -- the notebooks are the code, with the exception of the Ellipse Editor and Console Scripts. Currently this is an anonymous mirror of a non-anonymous repository of a very similar name.

Datasets

We weren't sure how to anonymously release a dataset on Zenodo. So for now, during the review process, we have datasets stored on Dropbox that are accessible via the get_data() function. Arguments to get_data are:

• 'cleaner' = Our Real dataset (clean-er than precleaned)
• 'preclean' = The aggregated output of 15 or more volunteers (per annotation) on Zooniverse
• 'spnet' = The formerly-private Real dataset from Hawley & Morrison's "ConvNets for Counting..."/"SPNet" paper. Released with the original authors' permission.
• 'cyclegan' = The 1200-image "CGSmall" subset of the public "SPNet Dataset Release" used for the SPNet paper. Note that the ring counts in this dataset are only integers.
• 'fake2' = We created a new 'fake' dataset like the one used on SPNet but without the former's limitation of integer-quantized ring counts. ...We didn't get a chance to CycleGAN it in time, but we plan to do so soon.

Pending favorable review, these datasets will be released on Zenodo with relevant authors' names.

Note: The above datasets are each approximately 2000 images and their corresponding labels, crops, etc. One dataset we will not be releasing is the full 14,000-frame unlabeled set of videos of the drum, from which a few of the paper graphs, i.e. those in Figure 4, are derived. These 14000 frames are being kept back as "proprietary" so that the authors can learn & publish about additional physics now that we have the tools to study this sytem.

Pretrained Model Weights

These too are stored on Dropbox for now, and the function get_checkpoint() is intended to load them as with get_data. We release weights for the real dataset tasks of

• bounding box detection
• counting rings in cropped images
• segmentation-regression
• segmentation (all one class)

Full Release Directory

The data and the weights both exist in Dropbox as sub-directories of a publicly-accessible espiownage-release directory, accessible via this link. After the anonymous review process, a redirect will be placed there to point to the full (author-named) release repository.