BloomLens: A Curated 23-Class Flower Image Dataset for Computer Vision and Deep Learning

BloomLens is a multi-class flower image dataset developed to support research in computer vision and deep learning, with a focus on robust flower species recognition under real-world conditions. The dataset contains 23 classes of commonly found ornamental and wild flowers, where each class is organized in a separate folder and labeled using both common names and scientific names.

The dataset includes a total of 8,925 images, collected in natural outdoor environments with realistic variations in lighting, viewpoint, distance, background, and occlusion. These real-world variations improve the robustness of machine learning and deep learning models, making BloomLens suitable for practical flower classification and benchmarking tasks.

Dataset Composition

Classes (23): Yellow Allamanda, Butterfly Pea, Paperflower, American Dwarf Heliconia, Crape Myrtle, Frangipani, Rose, Pink Kopsia, Crown-of-thorns, Singapore Daisy, Marigold, White Jasmine, Shoeblackplant, Thryallis, Philippine Ground Orchid, Madagascar Periwinkle, Jungle Geranium, Arrowleaf Sida, Pampas Grass, Bleeding-heart Vine, Sweet Autumn Clematis, Princess Flower, and Rose Cactus.

Total Images: 8,925

Label Format: Folder-wise labeling (one folder per class), with serial ID and class names defined using both common names and scientific names.

Collection and Diversity

BloomLens images were captured in real-world scenarios to ensure diversity and support model generalization. The dataset includes natural variations such as:

Different illumination conditions, including sunny, cloudy, and shadowed environments

Multiple viewing angles and flower orientations

Varying backgrounds, such as gardens, roadsides, and natural vegetation

Differences in flower size, growth stage, and partial occlusion

This diversity makes the dataset useful for training and evaluating deep learning models in realistic conditions rather than controlled laboratory settings.

Repository and Metadata

The dataset is prepared for upload to Zenodo as an open-access dataset. The repository includes:

Class-wise image folders for all 23 categories

A Class_list.csv file containing class IDs, common names, and scientific names

Clear documentation of folder structure and labeling conventions for easy use with machine learning and deep learning frameworks

Potential Applications

23-class flower image classification

Transfer learning and benchmarking using CNNs and Vision Transformers

Multi-class classification research in plant and flower recognition

Data augmentation and robustness experiments

Educational use for computer vision dataset practice and model training