Submission Guide

Welcome to the submission page for the Mouse vs AI: Robust Visual Foraging Challenge at NeurIPS 2025.

Submit via Codabench: https://www.codabench.org/competitions/9453/

Note: Submitting a model is sufficient to appear on the leaderboard. However, to be considered for inclusion in the NeurIPS workshop and the post-competition summary paper, top-performing teams must also provide a training pipeline. This can be submitted after the leaderboard deadline, but is required for final recognition.

1. Prepare Your Submission

  • Required: Include your trained model in .onnx format.
  • Optional but required for workshop inclusion: Submit a minimal training pipeline to allow retraining of your model (this can also be submitted retrospectively).
  • Optional: Add a short model description (.txt or .pdf) outlining your architecture, training procedure, and any notable design choices. You may also submit this description later using this Google Form.
  • Combine all files into a single .zip archive before submitting.

2. Submit Your Model

  1. Log in to your Codabench account.
  2. Navigate to the “My Submissions” tab.
  3. Click “Upload Submission”, select your .zip file, and fill in the required fields.

3. Evaluation and Leaderboard

  • Your model will be automatically evaluated on our standardized test environments.
  • Results will be posted on the public leaderboard after evaluation completes.
  • Final scores will be based on performance on a private held-out test set evaluated by the organizers.

Competition Rules

  • The competition is open to all researchers, students, and teams worldwide.
  • Participants may use any training method, architecture, or data augmentation approach.
  • Public datasets, pretrained models, and open-source libraries may be used for training.
  • All submissions must run entirely offline, without internet access.
  • No training is performed during evaluation—inference only.
  • Submissions must be compatible with the interface provided in the starter kit.
  • Metadata (e.g., configuration files or logs) can be included for reproducibility, but are not required.
  • Final evaluation and ranking will be performed on hidden test conditions.
  • Top teams will be invited to contribute short reports for the NeurIPS workshop and summary publication.

For questions or technical issues, please reach out via Discord