Rubrics
Model Performance
Performance on Hidden Test Set: The model will be evaluated against a hidden test set of images and metrics like mAP (mean Average Precision), Precision and Recall will be calculated.
Roboflow performance: The mAP, precision and recall achieved on Roboflow will also be taken into consideration.
Methods and Techniques
Data annotation methods: How did you annotate the provided data?
Data augmentation methods: What augmentation methods did you apply to the dataset and why?
Data pre-processing methods: What pre-processing methods did you apply to the dataset and why?
Any other innovative method used at any step of the hackathon. Did it affect the model performance?
Submission
Each team will make one submission through the provided Google Form: https://forms.gle/3hsc368Vc7MccZL3A
A PDF report should be uploaded along with the form submission, ensuring it clearly documents your approach and results. Some key points to include in the report include the methods and techniques outlined in the rubric, model performance, challenges faced, innovative approaches used, and any other aspect relevant to the project.
The Google Form requires you to enter: the link to your final trained model, its API key, and the class names that you defined in Roboflow. To obtain these:
Select your project from the Roboflow app home. Click on 'Deployments' in the left menu.
Click on 'Self-Hosted inference'. In the dialog, select your final model from the dropdown menu. Click on 'Native Python'.
Copy the model link and the API key (without the quotes) and paste it in the form.
Select your project from the Roboflow app home. Click on 'Settings' in the left menu.
Copy the CLASS NAME of both the class names exactly as you have defined them and paste it in the form. For example, in the below image the class names are ryegrass and wheat.
The goal is to not just achieve the highest metrics but also demonstrate your understanding and problem-solving approach. Good luck!