Submission Guide
Each submission must include:
- A trained model file (.pt, .ckpt, or .onnx)
- Optional README or config for reproducibility
Submissions are evaluated centrally by the organizers on a held-out test set with visual perturbations.
Submission Guide — v0.9 Beta
Scope — These instructions apply to the Windows
.exebeta. For the upcoming v1.0 Linux release the packaging format will stay identical, but the runner script will change.
1 · Submission checklist
| Item | Notes | |
|---|---|---|
| Trained model files | .pt or .onnx; see §2 for naming convention |
|
config.yaml |
Snapshot of your training hyper‑parameters | |
metadata.txt |
Team name, contact e‑mail, short description | |
| Zip archive ≤ 500 MB | 7‑zip or standard zip; no .rar |
2 · Folder structure & naming
Your archive must unpack into a single top‑level folder named after your run‑id:
MyAgent_Normal/ <‑ top‑level folder
├── model_head.pt <‑ weights file (PyTorch example)
├── config.yaml
└── metadata.txt
- Multiple networks? Package each network in its own folder and zip them together:
MyTeam_Submission_v2.zip
├── VisNet_Normal/
│ └── ...
└── ResNet_Perturb/
└── ...
3 · Send us your submission
- Upload your .zip archive (≤ 500 MB) to any cloud‑storage service you prefer—Google Drive, Dropbox, OneDrive, or WeTransfer all work.
- Set the link to “Anyone with the link can view”.
- Fill out the short Submission Form (takes < 1 min): https://forms.gle/COMP-Submission-Form
- You will receive an automatic confirmation e‑mail. We will download the archive, run the evaluation manually, and add your score to the public spreadsheet (updated every Monday & Thursday).
Submission limit — You may submit up to 3 archives per calendar week. Please wait for your previous score before sending the next one.