# SapBark-64: A dataset of bark images for 64 fruit-tree sapling classes

## TL;DR

SapBark-64 is a new dataset of bark images from 64 fruit-tree sapling classes, designed to support plant identification and horticultural analysis.

## Contribution

The paper introduces SapBark-64, a curated and publicly available dataset of fruit-tree sapling bark images with detailed metadata.

## Key findings

- SapBark-64 contains 5742 images of 64 fruit-tree sapling classes with standardized metadata.
- The dataset includes both raw and background-removed images, enabling fine-grained identification and trait-conditioned analysis.
- The dataset is publicly available under a permissive license with detailed documentation for reuse.

## Abstract

We present SapBark-64, a curated dataset of 5742 close-range bark images from 64 fruit-tree sapling classes (species/cultivar). Images were acquired in situ at three commercial nurseries in Trabzon (Türkiye) in 2025, targeting 1–2-year saplings routinely traded in nurseries. Photographs were captured with an iPhone 16 Pro Max at approximately 10 cm from the trunk under near-uniform illumination, using a white background to occlude scene clutter and preserve fine-scale morphology. For each class, a nursery label photo was recorded to support ground truth, and class-level characteristics were collected at the time of recording under expert supervision.

The repository is organized as two parallel image folders plus a structured metadata workbook: (i) raw images (JPG) and (ii) background-removed images (WebP) that mirror the same 64 class folders named by species/cultivar, enabling one-to-one pairing across versions; and (iii) an Excel (XLSX) metadata file list- ing standardized fields (family, scientific/common name, cultivar/variety, sapling height, trunk diameter, best planting season, growth rate, fruit-bearing age, average yield, production region, propagation method). This organization facilitates fine- grained identification and retrieval tasks and supports trait-conditioned analyses linking visual texture to horticultural attributes.

The dataset is publicly available in an open repository under a permissive license; acquisition conditions, directory layout, and the metadata schema are documented to enable unambiguous reuse.