2nd LPT Challenge Case 3 Datasets
The experimental dataset comprises images of Helium-filled soap bubbles within a Rayleigh-Bénard convection cell, captured by four (out of six) 5.5 MegaPixel SCMOS cameras (Bosbach et al. 2021, Godbersen et al. 2021, Weiss et al. 2024). The experiment was conducted at DLR Göttingen and consisted of an electrically heated aluminum plate and a water-perfused cooling plate, which allows for illumination from the top using an assortment of LED arrays (see Fig. 1). A total of six PCO edge 5.5 SCMOS cameras (2560 × 2160 pixels with a 6.5 µm pitch and an average magnification of 1.8942 px/mm) was installed and equipped with Zeiss Distagon 35 mm lenses. They were installed at the height of the upper cooling plate and slightly tilted downwards to view the full cell. Despite the strong reflection on the black bottom plate, bubbles are still visible in this region. The flow was recorded at a repetition rate of 30 Hz for over 30 minutes. The images captured by all six cameras over 54528 time-steps were used to reconstruct Lagrangian Particle Tracks using the DLR Shake-The-Box implementation; these results will serve as a Ground Truth for the evaluation of the LPT challenge results. For the challenge data, the four inner cameras of the in-line camera setup are selected (marked in red in Fig. 1, right), all viewing the entire volume within the cell. The omission of the two outer cameras severely aggravates the reconstruction problem, allowing the six-camera solution to be regarded as a Ground Truth.
The use of Helium-filled soap bubbles with limited lifetime leads to an ever-diminishing particle density over the measurement. This allows to extract short time-series of images at arbitrary particle image density. At the beginning, the six-camera evaluation is able to track around 550,000 bubbles, corresponding to over 0.13 ppp at an assumed active image area of around 4Mpix; at the end of the time-series 2385 tracked bubbles remain (see Fig. 2 right). Please note that the visible particle image density varies greatly over the image due to the circular shape of the sample, therefore the given ppp-values represent averages over the active image.
For the reconstructions with only four cameras we supply time-series of 50 images at 0.01, 0.04 and 0.07 ppp and 100 images at 0.1 ppp. Two-pulse-cases are available at 0.005, 0.015, 0.045, 0.06 and 0.075 ppp (see Table 1). The disjoint values for TR and TP are chosen to avoid an overlap of flow situations and images between the two modes. For each case, a sliding minimum image is provided for image preprocessing, created by taking for each pixel the minimum over +/- 10 images of the first image (see Fig. 2).
For calibration, two options are available – either a total of around 75.000 random points within a slightly enlarged reconstruction volume (‘LPT_CASE03_CalibPoints.txt’) or a version with around 14.000 points on several x-y-planes at different z-depths (‘LPT_CASE03_CalibPoints_on_grid.txt’). The point correspondences were created using the fully calibrated two-plane camera model (initial plate calibration, volume self-calibration and finally a B-spline based 2D correction field which accounts for the curvature of the plexiglass) that was used within the six-camera STB evaluation. Participants can either just use these points to calibrate their camera models or additionally perform a Volume-Self-Calibration using the supplied time-series.
The coordinate system is approximately centered in the sample, with X, Y and Z within (-550 to +550, -610 to 500, -550 to +550,) mm. The dynamic range of the velocity is rather high, with low particle shifts in some regions, while the highest velocities can reach around 0.35 m/s, corresponding to a particle shift of around 20 pixels (these values have not been checked for all data points, so they should not be used as absolute limits).
Requested output and formatting rules
Participants can choose whether they want to process the TR or the TP data, or both. For TR, results have to be submitted at least up to ppp = 0.07 and for TP up to ppp = 0.045. Processing the higher seeded cases is of course encouraged.
Common to all cases: submission form and zip archive
Clicking on "Submit TP results" or on "Submit TR results" will open a form allowing you to upload a file containing your results. In either case, this file should be a zip archive (.zip extension), containing directly (no folder) one result file per seeding density, each being an ASCII file (see further precisions regarding file naming and format in the paragraphs below).
The upload form should also be filled with the following information, which is used for result presentation only if you choose to publish your result based on the evaluation sent by email:
- the short name or acronym of the algorithm used for data processing (should match with the prefix name of your result files, see below). This name should have a length of 24 chars maximum, and contain no space. Authorized chars are alphanumeric chars and one of the following: [ ] _ - @
- the full name of the algorithm. This name should have a length of 100 chars maximum, and should not contain commas.
- Optionally, you may provide the URL address of the publication (if any) about your algorithm, or of a webpage describing it
- If needed (e.g. in the situation where the algorithm is due to a collaboration between two or more institutions), you can correct the name of the Institution which will appear in the public result tables (set by default to your institution). For this field, the maximum length is of 24 chars and commas are not allowed.
TR case
For TR, the tracked particles for time step 25 (images ‘_I0024’, ppp ≤ 0.07) or 50 (images ‘_I0049’, ppp = 0.1) need to be supplied for each processed seeding density in the following format: ASCII-file with first line X Y Z Xfit Yfit Zfit VX VY VZ AX AY AZ, followed by one line for each tracked particle with raw position (in mm), fitted position (temporal fit of participants choice to dampen position noise and to determine velocity and acceleration; in mm), the velocities VX, VY, VZ (in m/s) and the accelerations AX, AY, AZ (in m/s2), all calculated at time step 25 or 50 depending on the seeding density of the case. The naming of the files should follow the convention “ZZZZ_LPT_CASE03_TR_ppp_0_AAA_ PartFieldNN.dat”, with ZZZZ being the algorithm acronym or short name as provided on the submission form (see previous paragraph), AAA fractional ppp and NN the time-step index (24 or 49).
TP case
For TP, the single computed flow field needs to be supplied for each processed seeding density in the following format: ASCII-file with first line X0 Y0 Z0 X1 Y1 Z1, followed by one line for each particle with the two measured particle positions X0, Y0, Z0, and X1, Y1, Z1 (in mm) for time steps 0 and 1 of all identifies two-pulse track. The naming should be “ZZZZ_LPT_CASE03_TP_ppp_0_AAA_ PartField00.dat” with ZZZZ being the algorithm acronym or short name as provided on the submission form (see previous paragraph), AAA fractional ppp.
Contact in case of any questions: Daniel Schanz, DLR: daniel.schanz@dlr.de
Literature
Bosbach J, Schanz D, Godbersen P, Schröder A (2021) Spatially and temporally resolved measurements of turbulent Rayleigh-Bénard convection by Lagrangian particle tracking of long-lived helium-filled soap bubbles, 14th International Symposium on Particle Image Velocimetry
Godbersen P, Bosbach J, Schanz D, Schröder A (2021). Beauty of turbulent convection: a particle tracking endeavor. Physical Review Fluids, 6(11), 110509.
Weiss, S., Schanz, D., Erdogdu, A. O., Schröder, A., & Bosbach, J. (2024). On Lagrangian properties of turbulent Rayleigh–Bénard convection. Journal of Fluid Mechanics, 999, A90.