Performance of object detector use case is computed using mAP (mean Average Precision) : https://medium.com/@jonathan_hui/map-mean-average-precision-for-object-detection-45c121a31173
This metric is computed on up to 5 differents models trained on the Data Set.
Please note that the performance spread can be important between first and last models.
A random sample of images are displayed with:
- In blue the true bounding box, as supplied by the user
- In orange the predicted bounding box on predicted by Prevision.io on cross validation
Only box with a probability > 10% are displayed in orange.
If none are display, that means that there is no significant detection.
Prediction works like other image Data Set.
Once done, you can directly check results with probability associated with each bounding box by clicking the see predictions link: