SorpVision Dataset

This dataset contains both the original and resized materials used for our sorptivity analysis experiments. The real dataset was originally captured at 640×480 (VGA) resolution, while the synthetic dataset was natively generated at 448×448 pixels. To ensure full transparency and reproducibility, we provide all data in three main folders:

videosThis folder contains the original VGA “speed-up” videos recorded at 640×480 pixels. These serve as the unmodified source material for the real dataset.

video frames – annotated by 4 peopleThis folder contains the extracted video frames, each at 640×480 resolution, with pixel-level annotations provided by four independent annotators. These frames capture the same details found in the original videos, allowing for direct evaluation and comparison of annotation quality.

datasetThis folder contains the 448×448 images and their corresponding masks used in our final training and analysis. The real images in this folder have been resized down from 640×480, while the synthetic images were originally created at 448×448.

By offering both the native-resolution and resized data, we hope to facilitate a wide range of research objectives. Users may employ the original files for detailed studies requiring maximum fidelity, or use the 448×448 dataset for consistency with our published results. In all cases, the complete collection aims to support reproducibility and flexibility in experimentation.