Training metrics

To view your project’s training metrics, go to Neural Network and select the training model. Select Download to download the training metrics.

Instance segmentation for Pixel Projects and Vector Projects

MetricDescriptionPossible range
Total lossThe current value of the loss function as described in Mask RCNN, a paper by Kalming He et. al.Any range
Mask lossThe current value of the component of the total loss function that is responsible for the mask head.Any range
Classes lossThe current value of the component of the total loss function that is responsible for the classes head.Any range
Bounding box lossThe current value of the component of the total loss function that is responsible for the bounding box head.Any range
Bounding box mAPThe mean average precision of the bounding boxes over all the training classes evaluated on the validation set.0-100
Bounding box mAP at IoU=0.50The mean average precision of bounding boxes at IoU=0.5 evaluated on the validation set.0-100 (per each IoU)
Bounding box mAP at IoU=0.75The mean average precision of bounding boxes at IoU=0.75 evaluated on the validation set.0-100 (per each IoU)
Bounding box mAP forThe average precision of bounding boxes found for particular class, i.e., how precise the model was for a given class. It is reported for all the training classes.0-100
Segmentation mAPThe mean average precision of the segmentation over all the training classes.0-100
Segmentation mAP at IoU=0.50The mean average precision of segmentation at IoU=0.50-100
Segmentation mAP at IoU=0.75The mean average precision of segmentation at IoU=0.750-100
Segmentation mAP forThe average precision of the segmentation found for a particular class, i.e., how precise the model was for a given class.0-100
ETAEstimated time of arrival

Object detection for Vector Projects

MetricDescriptionPossible range
Total lossThe current value of the loss function as described in Mask RCNN, a paper by Kalming He et. al.Any range
Classes lossThe current value of the component of the total loss function that is responsible for the classes head.Any range
Bounding box lossThe current value of the component of the total loss function that is responsible for the bounding box head.Any range
Bounding box mAPThe mean average precision of the bounding boxes over all the training classes evaluated on the validation set.0-100
Bounding box mAP at IoU=0.50The mean average precision of the bounding boxes over all the training classes at IoU=0.50 evaluated on the validation set.0-100 (per each IoU)
Bounding box mAP at IoU=0.75The mean average precision of the bounding boxes over all the training classes at IoU=0.75 evaluated on the validation set.0-100 (per each IoU)
Bounding box mAP forThe average precision of the bounding boxes found for a particular class, i.e., how precise the model was for a given class. It is reported for all the training classes.0-100
ETAEstimated time of arrival

Semantic Segmentation for Pixel Projects

MetricDescriptionPossible range
Semantic segmentation lossPixel-wise cross entropy lossAny value
ETAEstimated time of arrivalAny value
mIOUMean intersection over union over all the classes0-100
mACCMean accuracy over all the classes0-100
ACCAccuracy per class0-100