Evaluation

The design is evaluated based on accuracy and throughput.

ACCURACY

Accuracy is measured using the F1-score, where:

F1-score = (2 * precision * recall) / (precision + recall)

precision = global_true_positives / (global_true_positives + global_false_positives)

recall = global_true_positives / (global_true_positives + global_false_negatives)

True positives are when a reported object matches the type and location (IoU > 0.5) of an object from the golden data.

mIoU is used to measure the segmentation performance.

The minimum accuracy should be TBD, otherwise a penalty is applied.

THROUGHPUT

Throughput is measured in FPS. The time to read images from the SD card is not included.

The design should achieve at least 5 FPS (subject to change), otherwise a penatly is applied.

SCORING FUNCTION

The score for a team is calculated as follows (Subject to change):

team score = F1-score^2 * fps

team score = (F1-score^2 + mIoU^2) * fps

Note that for fairness, all valid submissions must meet the requirements: F1 >= 0.25, mIoU >= 0.15.

Scoring Script

A score_miou.py script is provided that will provide f1-score, mIoU and fps.