---
license: mit
task_categories:
- image-classification
- object-detection
tags:
- biology
- plants
- succulent
- unsupervised-learning
- dinov2
dataset_info:
  features:
  - name: image
    dtype: image
  - name: label
    dtype:
      class_label:
        names:
          0: class_000
          1: class_001
          2: class_002
          3: class_003
          4: class_004
          5: class_005
          6: class_006
          7: class_007
          8: class_008
          9: class_009
          10: class_010
          11: class_011
          12: class_012
          13: class_013
          14: class_014
pretty_name: High-Quality Clustered Succulent Dataset
size_categories:
- 1K<n<10K
---

# Succulent Vision Dataset

## Introduction
This dataset contains **1,000+ high-quality segmented images** of succulents. 
The original photos were taken in complex cluster environments, and individual succulent plants were precisely segmented using **SAM (Segment Anything Model)**.

## Processing Pipeline
To provide structured data, the images have been automatically categorized using an advanced unsupervised pipeline:
- **Feature Extraction**: [DINOv2](https://github.com/facebookresearch/dinov2) (Vision Transformer) for semantic shape features.
- **Color Analysis**: **HSV Histogram** analysis to prioritize botanical color variations.
- **Dimensionality Reduction**: **UMAP** for manifold learning.
- **Clustering**: **Agglomerative Hierarchical Clustering** to ensure morphological consistency.

## Dataset Structure
The dataset is organized into folders representing different species/morphological groups:
- `class_000/`: Echeveria-like structures (green)
- `class_001/`: Cabbage-like wrinkled leaves
- ... and more.

## Usage
Perfect for:
- Training **Conditional DDPM** or **GANs** for plant generation.
- Fine-tuning image classification models for botany.
- Testing unsupervised clustering algorithms.

**Maintained by**: [HaiPenglai](https://huggingface.co/HaiPenglai)