Registration Required

All participants must register to join the challenge. After registration, you'll receive essential credentials and updates.
👉 Register here

Track Your Progress

Submit your interim results through our Results Portal to view your temporary ranking. A password is required (you will receive it after registration).
👉 Submit results

Submission Format:

Files must be in COCO format (JSON file following the COCO evaluation schema).
👉 Sample Submission File

🏆 Final Submission Requirements

Your GitHub repository MUST follow this exact structure:

To ensure consistency and fairness in the evaluation process, top 10 teams have highest performance submission must include a GitHub repository name as TeamName_CADOT_Challenge structured as below, containing the specified components.

• README.md: Must contain:
  • Hardware/software requirements (e.g., CUDA version, GPU memory).
  • Step-by-step execution commands with example parameters.
  • Validation scores achieved during development.
  • Dependency installation instructions (preferably using virtual environment).
• requirements.txt: Frozen dependencies with exact versions (e.g., torch==2.1.0).
• data/: Empty directories with instructions.md explaining the directory structure for generated images (if using augmentation).
• models/:
  • Pre-trained weights (.pth, .pt, .h5).
  • Architecture file with all layer details.
  • Config file with readable hyperparameters.
• scripts/:
  • Training script with full reproducibility.
  • Inference script producing COCO-format JSON.
  • Separate module for data augmentation.
  • Evaluation script calculating all metrics.
• results/:
  • Final predictions.json for the test set.
  • metrics.csv with per-class AP/AR metrics.
• reports/:
  • 5-page PDF report including:
  • Model architecture diagram
  • Training convergence plots
  • Error analysis with confusion matrix

Technical Requirements – Critical Aspects:

• Reproducibility: Must achieve ±1% mAP@50 variation from reported scores when re-run by organizers.
• Annotation Compliance: Predictions.json must use exact COCO-format with official class IDs.

Forbidden Practices:

• Manual annotation of any generated images.
• External data not documented in report.