A multi-source dataset of gas–liquid bubble flows with annotated synthetic and simulated data and real-world imaging

DATASET: bubbles_v0_8_0 2 Short Description This dataset comprises a series of temporal snapshots capturing various configurations of both simulated and real air-bubble injections into the liquid phase. The bubbles rise vertically within a cylindrical glass tube with a diameter of 15 cm and a height of 80 cm. The water level is approximately 35 cm in the case of a simulated and modeled water column and approximately 70 cm in the case of an real tank. The air inlet is located at the bottom of the tube, and its size and flow rate are defined separately for each case. Simulated cases were generated using OpenFOAM and rendered using a ray tracer. Modeled cases were rendered using the same custom ray tracer. The experimental cases were captured using a real water tank and a digital camera. Individual cases described below shows the behavior of separated, interacting, clustered, and turbulent bubble flows. Dataset Structure The dataset is divided into four cases. (a) The first case represents four different regimes of artificially modeled bubbles (files mR?_*.jpg, where R is 3, 4, 5, or 6) captured from two different perspectives (files m?P_*.jpg, where P is 0 or 2). Corresponding labels in YOLO format for detection and segmentation tasks are also generated for these images. (b) The second case is structured in a similar manner, representing a physical simulation of bubble flow through a tube carried out with OpenFOAM. The images are generated from two perspectives (files s2P?_*.jpg, where P is 0 or 2) for three different inlet air velocities (files s2?R_*.jpg, where R is 5, 6, or 8). (c) The third case is a merged variant of the two preceding cases (a-b), making it the most versatile scenario for machine learning applications. (d) The fourth and final case comprises images capturing real bubbly flow. Each time stamp represents a time interval of 0.01 seconds. All rendered images have a resolution of 1000 x 1500 pixels and a field of view of 24 degrees in the horizontal plane of the camera, which corresponds to real images taken by a camera with a focal length of 85 mm and a full-frame sensor. (a) Modeled Case (bubbles/modeled, 4588 train images) ``` bubbles/ ├── modeled/ │ ├── test/ (the same applies to the train/ and val/ directories) │ │ ├── images/ │ │ │ └── *.jpg │ │ ├── labels_det/ (labels for detection task) │ │ │ └── *.txt │ │ ├── labels_seg/ (labels for segmentation task) │ │ │ └── *.txt │ │ └── labels/ (symlink to the labels_det/ or labels_seg/ directory, depending on the required task) │ ├── ... │ ├── ... │ ├── class.txt │ └── bubbles_dataset.yaml (change the path attribute if necessary) ``` Note that the labels/ directories in the test/, train/, and val/ directories are created as symlinks to the corresponding directories depending on the task being performed. Make sure you have set the labels/ directories appropriately. (b) Simulated Case (bubbles/simulated, 1988 train images) The same directory structure as in the case of the modeled dataset also applies to the simulated dataset. Also make sure you have set the labels/ directories appropriately. ``` bubbles/ ├── simulated/ │ ├── test/ (symlink to ../modeled/test) │ ├── ... │ ├── ... │ ├── class.txt │ └── bubbles_dataset.yaml (change the path attribute if necessary) ``` (c) Merged Case (bubbles/merged, 6576 train images) This directory contains the combined modeled and simulated datasets. It is the primary directory intended for training. Note that the contents of both datasets are referenced via symlinks. ``` ├── merged/ │ ├── images/ │ │ ├── test/ │ │ │ └── real/ (symlink to ../../../modeled/test/images) │ │ ├── train/ │ │ │ ├── modeled/ (symlink to ../../../modeled/train/images) │ │ │ └── simulated/ (symlink to ../../../simulated/train/images) │ │ └── val/ │ │ ├── modeled/ (symlink to ../../../modeled/val/images) │ │ └── simulated/ (symlink to ../../../simulated/val/images) │ ├── labels/ │ │ ├── test/ │ │ │ └── real/ (symlink to ../../../modeled/test/labels) │ │ ├── train/ │ │ │ ├── modeled/ (symlink to ../../../modeled/train/labels) │ │ │ └── simulated/ (symlink to ../../../simulated/train/labels) │ │ └── val/ │ │ ├── modeled/ (symlink to ../../../modeled/val/labels) │ │ └── simulated/ (symlink to ../../../simulated/val/labels) │ ├── class.txt │ └── bubbles_dataset.yaml (change the path attribute if necessary) └── README.md (this file) ``` (d) Real Case (bubbles/real, 1970 images) This directory contains four real variants of air flow through a cylinder captured by a real camera. ``` └── real/ ├── low/ │ └── *.jpg ├── mid/ │ └── *.jpg ├── mid_high/ │ └── *.jpg └── high/ └── *.jpg ```