Average Overlap (AO) is the average overlap of the sequence of IoU (Intersection over Union) of the pixels between tracking result and groundtruth bounding box.
where
- K-Means vs. Fuzzy C-Means for segmentation of orchid flowers, III. RESULT AND EVALUATION [paper]
- https://github.com/got-10k/toolkit/blob/v0.1.3/got10k/experiments/got10k.py#L265
Area Under Curve (AUC) is the average of the success rates corresponding to the sampled overlap thresholds. The AO is recently proved to be equivalent to the AUC. AUC value is usually used to ranking the trackers in success plot
where,
- GOT-10k: A large high-diversity benchmark for generic object tracking in the wild, 4.2 Evaluation methodology section [paper]
- https://github.com/got-10k/toolkit/blob/956e7286fdf209cbb125adac9a46376bd8297ffb/got10k/experiments/lasot.py#L144
P stands for precision score. Usually measured as the distance in pixels between the centers
where
- TrackingNet: A large-scale dataset and benchmark for object tracking in the wild,3.4 Evaluation section [paper]
- https://github.com/got-10k/toolkit/blob/v0.1.3/got10k/utils/metrics.py#L7
where,
then,
- TrackingNet: A large-scale dataset and benchmark for object tracking in the wild,3.4 Evaluation section [paper]
- https://github.com/got-10k/toolkit/blob/956e7286fdf209cbb125adac9a46376bd8297ffb/got10k/utils/metrics.py#L22
# This is an example for GOT-10k, TrackingNet, and LaSOT dataset
pip install numpy pandas opencv-python tqdm
python metric.py
--- Evaluate for GOT-10k ---
Average Overlap (AO): 28.40 %
Success 0.5 (SR0.5): 22.12 %
Success 0.75 (SR0.75): 16.81 %
--- Evaluate for TrackingNet & LaSOT ---
Success score (AUC): 28.78 %
Precision score (P): 23.89 %
NPrecision score (P_norm): 21.90 %
- [1] GOT-10k: A large high-diversity benchmark for generic object tracking in the wild [paper]
- [2] TrackingNet: A large-scale dataset and benchmark for object tracking in the wild [paper]
- [3] LaSOT: A high-quality benchmark for large-scale single object tracking [paper]
- [4] Object tracking benchmark [paper]
- [5] Visual object tracking performance measures revisited [paper]
- [6] K-Means vs. Fuzzy C-Means for segmentation of orchid flowers [paper]
- [7] Robust visual tracking with reliable object information and Kalman filter [paper]
- [8] Online Object Tracking: A Benchmark [paper]
- [9] Intersection over Union (IoU) for object detection [link]
- [10] Understanding AUC-ROC Curve [link]