2nd LPT Challenge Case 2 TP Results

Second LPT Challenge Case 2: Two-Pulse results

This page presents published results for the TP (Two-Pulse) case of the 2nd LPT Challenge Case 3. The link for result submission, as well as details about the test case and rules for result formatting are given there.

Tables can be ordered according to each of the quantities reported. Clicking on an algorithm name in a table will open a page gathering its result, together with information on the submitter.

Definition of performance metrics

Evaluation pertains to both particle localization, at the two pulse times t0 and t1, and estimation of the displacement vectors. For this specific test case, where a subset of the particles are viewed by only 2 or 3 cameras at low I/U ratio, quantities for performance evaluation are estimated on the set of particles which are viewed by all 4 cameras. For either particles or vectors, we count:

• a True Positive (TP) if a ground truth (GT) particle is found in the neighborhood of a detection (maximum componentwise distance less than 1 voxel). For vectors, both detections should be TPs and associated to the same GT particle.
• a False Negative (FN, a.k.a. missed particle) if there is no detection in the neighborhood of a GT particle. A FN vector is a GT vector that is not associated to a TP vector.
• a False Positive (FP, a.k.a. ghost particle/vector) particle/vector when it is not a TP particle/vector.

Precision and Recall are derived quantities reflecting the detection performance, defined as:

Root-mean-square (rms) error on particle position or displacement are defined by only considering the TP within the result. 

In the tables, they are given in voxel, with a voxel size here equal to 30.5 µm; figures also show these errors in mm. Conversion from displacement to velocities can be done straightforwardly by considering the inter-frame time, set to 40 µs; for reference, the free-stream velocity is here equal to 10 m/s.

Below the tables, we provide plots comparing some of these quantities among participants, as well as displacement error slices for selected test points.