If you use this dataset please cite the DeepCS-TRD manuscript. Method Image and annotations can be inspected using the TRAS tool developed during the project. 1) Open TRAS 2) Open Image 3) File>Load Annotations Harvest Site The samples were collected in Curticeiras, Department of Rivera, Uruguay, during 2022 (18 samples) and 2023 (35 samples) as part of a harvesting operation managed by a local lumber company. The sampling locations correspond to the following coordinates: 31°17'45" S, 55°41'42" W; 31°10'4" S, 55°39'42" W (18 samples); and 31°01'11.37" S, 55°31'55.81" W (35 samples). In total, 53 cross-section samples were obtained. The site is classified under the CONEAT 7.31 group and is characterized by being located on the Tacuarembó Sandstone geological formation. It presents slopes ranging from 6% to 10%, very good drainage, and deep soils reaching up to 180–200 cm in the B horizon. Image Acquisition The UruDendro1 collection was expanded with 53 new images captured under laboratory conditions using an iPhone 6S (12 MP RGB sensor) positioned at a distance of 43 to 51 cm from the wood cross-section, and illuminated with a 35 W LED ring light. The surface preparation of the samples varied: some were cut using a chainsaw, others were smoothed with a handheld planer, and several were further polished with a rotary sander. All images were acquired indoors. The image widths range from a minimum of 2267 pixels to a maximum of 3024 pixels. Sample codes follow the structure: XY Where X denotes the diameter class to which the disc belongs (A, B, or C; see table), and Y corresponds to the sample identifier. Class A Class B Class C 35-40 cm 40.1-45 cm 45.1-50 cm Annotation Details The software tool used to generate the annotations was Labelme , which produces closed polygons defined by a set of vertices. We worked with two types of annotations:those corresponding to annual ring boundaries or earlywood–latewood boundaries, and those corresponding to reaction wood.

File structure In general, the file structure for each collection goes as follows: images/ - Folder with the image samples images_no_background/ - Folder with the images after the background has been removed annotations/– annual_rings/ - Annual ring annotations– earlywood_latewood/ - Earlywood-Latewood boundary annotations– reaction_wood/ - Reaction wood annotations pith_location.txt - file with the pith pixel location for each image Where each sample has a unique code. For example, sample F02c has five different files: images/F02c.jpg images_no_background/F02c.jpg annotations/annual_rings/F02c.json annotations/earlywood-latewood/F02c.json annotations/reaction_wood/F02c.json