🏆 Leaderboard
Challenge Overview
The current goal of automated proofreading is a difficult challenge involving many distinct, intermediate skills. In the interest of facilitating fully autonomous proofreading, we have prepared a series of intermediate challenges that incorporate essential skills for proofreading.
For each sub-challenge, training can be performed on the publicly available data hosted on Flywire. Submissions are currently being evaluated manually; we hope to develop an automated framework for leaderboard submission soon. Please reach out to raphael_levisse@berkeley.edu, kp0374@princeton.edu and dl2635@princeton.edu for additional information.
Submissions will be evaluated on private data not made available to the public.
Sub-Challenge 1: Navigation
Sub-Challenge 1 is a navigational task: to achieve the highest z-position for a given neuron. Your agent, when initialized at an arbitrary XYZ-position on a random neuron, should be able to navigate (via clicking, keyboard interactions, etc.) to the top z-position of the neuron.
Evaluation Criteria
- Rotational invariance
- Can the highest z-position be obtained regardless of initial angular orientation of the z-axis?
- Depth understanding
- Can your agent successfully identify the highest z-position in visually ambiguous situations?
- Speed
- How efficient is your agent at navigating?
Demonstrations
An example of reaching highest z-position.