Overview: The dataset VDS2Raw (v2) (Vessel Detection from Sentinel-2 Raw) is a collection of raw granules derived from Sentinel-2 products, specifically focusing on vessels in Danish coastal areas. The dataset is utilized for ship detection and analysis purposes. Disclaimer**: This new version of the dataset, VDS2Raw (v2), expands on the original dataset by incorporating the B8 band and AIS (Automatic Identification System) records, thereby enhancing the original dataset. Dataset Construction: 1. Reference Dataset: - The reference dataset was based on the work of Ruiloba et al. (2020), using Sentinel-2 L1C tiles acquired in 2019. - This dataset contained 1426 ship images with 24x24 pixel dimensions. 2. L0 Granule Selection: - A polygon encompassing the region of interest (ROI) and a specific date range was defined to download the relevant L0 granules. - A total of 390 L0 granules were retrieved where the reference band (B02) intersected the ROI during the time frame of 2019. A margin was applied to ensure the inclusion of all relevant bands, including the new B8 band. - The granules were decompressed to obtain raw, processable data, and co-registered using ESA tool PyRaws). 3. Labeling Process: - Ships were manually annotated with bounding boxes using bands B02, B03, B04, and the newly included B8 band. - A coarse spatial coregistration technique (Meoni et al., 2023) was used to aid in manual labelling. - Any missing elements due to the coregistration process were filled with adjacent granule pixels when possible, and areas with missing pixels were cropped. This resulted in granules with varying pixel areas. 4. AIS Data: - In addition to the optical data from the Sentinel-2 bands, AIS (Automatic Identification System) information was included for each granule to provide metadata for vessel identification and validation of ship detection. 5. Subset Selection: - From the original 390 granules, a subset of 166 was selected and divided into training (105), validation (27), and test (34) sets. - The number of ship annotations was 483 in the training set, 119 in the validation set, and 93 in the test set. 6. Granule Statistics: - The average pixel area of the granules was 2588.971 by 1669.44 pixels. - The mean annotation area was 13.59 square pixels by 15.67 square pixels.