Model Evaluation#
Evaluate model predictions against ground truth. Compute mAP, precision, recall, confusion matrices, and analyze failure cases.
Install#
curl -sL skil.sh | sh -s -- voxel51/fiftyone-skills
When prompted, select fiftyone-model-evaluation from the menu.
Requirements#
FiftyOne MCP Server (optional, recommended for operator-based evaluation)
@voxel51/evaluationplugin (optional, for operator-based evaluation)
Usage#
Load a dataset with both predictions and ground truth, then ask your AI assistant:
"Evaluate my YOLO predictions against ground truth"
"Compute mAP for my detection model"
"Show me the confusion matrix and top failure cases"
The skill checks for the required fields, runs the evaluation protocol, and opens the results in the App so you can drill into TP/FP/FN samples.
Example#
Then ask your assistant:
"Evaluate the predictions field against ground_truth in the quickstart dataset"