Spruce Lamellae Dataset

Dataset description:
This dataset consists of 2525 spruce lamellae, each photographed from two sides, resulting in a total of 5050 images of spruce boards.

For each board, the following ground-truth information is provided:

• A manually assigned quality class (categorical label) for each image

• The Modulus of Elasticity (MOE) and Modulus of Rupture (MOR) measured via a standardized 4-point bending test

This record includes:

• All board images (organized per board and side). Due to size restrictions, each image was converted from the original BMP format used in the paper to a lossless PNG format. The repository contains 10 batches of images (505 images per batch), named part_1.zip through part_10.zip.

• A CSV file (dataset_spruce.csv) containing the metadata: quality class, MOE, MOR, and predefined train/test split.

• defect_detection_dataset.zip provides image crops and YOLO-format defect annotations for instance segmentation model training. See https://github.com/Juliaachatz/LamellaVision for details.

File naming convention

Images follow the naming pattern: jjjj_mm_dd-XZ1.Z2.S.bmp, where:

• X = optional identifier (single letter)

• Z1 = batch number

• Z2 = sample number

• S = board side (1 = front, 2 = back)

In the original paper, we used .bmp files. To reduce the data load, we converted all images to png here.

Metadata file (CSV)

The CSV file contains the following fields for each sample:

• imagename (as described above)

• sample name (XZ1.Z2.S)

• quality class (1=Standard, 2=Industy, 3=Substandard quality)

• MOE (Modulus of Elasticity)

• MOR (Modulus of Rupture)

• density

• split (train/test)

The corresponding code is available in the following repository: https://github.com/Juliaachatz/LamellaVision