The dataset for Quantitative Evaluation of Apple Rotation and Surface Presentation on a Screw Conveyor for Full-Surface Imaging

Dataset Overview This dataset contains 1,421 annotated images of apples whose surfaces were painted with patches in 17 distinct colors. The colored patches serve as visual markers for investigating apple rotation behavior. Segmenting and tracking these regions across image sequences enables changes in the visible painted regions to be examined as apples rotate. This release provides annotated image data prepared for training color-region segmentation models. All images are stored in BMP format with uniform dimensions of 1600 × 2848 pixels (width × height). The release includes original polygon annotations created using LabelMe and YOLO-compatible TXT annotations converted from the JSON files. The segmentation models trained using these data are intended to support subsequent color-region tracking and analysis of apple rotation behavior. Dataset Organization The data are organized into three directories: Directory Contents Number of files labeled data/ BMP images corresponding to the annotations 1,421 json/ Original LabelMe polygon annotations in JSON format 1,421 txt/ YOLO-compatible annotations converted from the JSON files 1,421 The images and the two annotation formats correspond one-to-one. Each image has an associated JSON annotation file and a corresponding TXT annotation file. The JSON and TXT files provide two representations of the annotations for the same images. Dataset Summary Attribute Value Annotated images 1,421 Painted color categories 17 Image format BMP (.bmp) Image dimensions (width × height) 1600 × 2848 pixels, consistent across all images Reported BMP bit depth 32 bits per pixel Runs per apple at each speed Two: one in lane 1 and one in lane 2 Annotation software LabelMe Original annotation representation Polygons stored in JSON files Converted annotation representation YOLO-compatible TXT files Main application Training color-region segmentation models to support apple rotation analysis Approximate uncompressed size 28 GB Compressed archive size 7.39 GB Acquisition Runs Each apple was tested twice at each speed, once in lane 1 and once in lane 2. A test identifier denotes an experimental run, while a frame index identifies an image within that run. Multiple runs can therefore contain observations of the same physical apple. Research Purpose and Visual Markers The painted regions provide distinguishable surface markers for examining how different parts of an apple become visible during rotation. The segmentation task delineates the visible color regions in individual images. These predictions can subsequently be used in a tracking workflow to study changes in region visibility and appearance over time. The 17 color categories describe the painted visual markers used in the rotation study. Annotation Structure LabelMe JSON Annotations The json/ directory contains the original annotations created using LabelMe. The visible painted color regions were delineated using polygons. Each polygon describes the boundary of an annotated color region and is associated with its color category. Converted YOLO TXT Annotations The txt/ directory contains annotations converted from the original JSON files for use in YOLO training workflows. Each TXT file corresponds to the same image as its associated JSON file. File Naming and Correspondence Filenames identify the experimental run and the image index within that run. Some filenames additionally include a prefix describing acquisition conditions. For example: V4S2-HFYU_test-7_134.bmp Filename component Meaning V4 Variety identifier; S2 Operating speed code; test-7 Experimental run identifier 7; in this example, it corresponds to the third apple imaged in lane 2 134 The 134th image within the indicated run .bmp BMP image format The prefix V4S2-HFYU_ is not present in all filenames. The meanings of the trailing test identifier and image index remain the same when this prefix is omitted. The supplied file information also shows a filename without this prefix and without the hyphen after test, such as test1_1.bmp. Images, JSON annotations, and TXT annotations are paired one-to-one throughout the dataset. Users should preserve these pairings when reorganizing files or constructing data partitions. Availability of the Complete Dataset This release provides the annotated training subset. The complete dataset comprises approximately 2 TB of data. Researchers interested in accessing the complete dataset may contact the authors to request access. Usage Notes This release contains the annotated image collection prepared for model training. No predefined training, validation, or test split is supplied. Users should construct partitions appropriate to their intended experiments. Users training YOLO models should organize the images and TXT annotations according to the requirements of their selected training implementation and provide the appropriate dataset configuration and class mapping.