/
Extra credit 1 -- Machine Learning Performance Metrics

Extra credit 1 -- Machine Learning Performance Metrics

We used Google Collab to download image and label data from Labelbox, create XML labels, and train a Faster R-CNN model to detect plants. We did not, however, evaluate the model’s performance. This assignment is designed to expose you to ML performance metrics such as precision and accuracy, which must be reported alongside your trained model. You can earn up to 1.5x bonus points of a full homework assignment! 



Instructions

  1. Using the notebook we used during our in-class workshop, quantify the number of true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN). You will need (1) the model bounding box predictions and (2) test set ground-truth labels. This step can be done either manually or programmatically. 

    1. Article: Evaluating performance of an object detection model | by Renu Khandelwal | Towards Data Science

  2. Use TP, TN, FP, and FN counts to calculate the precision, accuracy, and F1-score for our trained model. 

    1. Article: See link above.

  3. Provide a table of performance metrics and a conclusion in your report.

    1. Would you trust this model if deployed in a real-world application?

    2. What may have gone wrong during preparation/training? 

    3. Should we go back to Labelbox and filter our labels? 

  4. Post the table and response onto the Wiki by September 23rd. 



Optional resources