---
license: mit

pretty_name: Foliagen
tags:
  - agriculture
  - plant-disease
  - image-generation
  - image-classification
  - synthetic-data
  - soybean
  - tomato
---

# Foliagen

Foliagen is a framework for creating realistic foliage images from high-resolution images of individual leaves. The generated dataset is intended to support research in foliar disease classification.
The corresponding paper has been published in IEEE Access. [Foliagen](https://ieeexplore.ieee.org/abstract/document/11666865)

## Features

- Generates foliage images from individual leaf images
- Supports healthy and diseased soybean leaves
- Places leaves on field-background images to improve visual realism
- Provides configurable foliage and plant-generation settings
- Includes classifier code for reproducing the experiments reported in the associated paper


## Input Data

### Individual leaf images

Individual leaf images are the primary inputs to the foliage generator. Each image must have a transparent background.

The current soybean configuration supports the following categories:

- Frogeye leaf spot
- Bacterial blight
- Rust
- Cercospora leaf blight
- Downy mildew
- Mosaic virus
- Potassium deficiency
- Sudden death syndrome
- Target spot
- Healthy

### Background images

Background images are used to emulate the field environment in which plants grow. Natural field backgrounds are recommended because plain backgrounds generally produce less realistic foliage images.

The background images used in the original experiments were obtained from [Freepik](https://www.freepik.com/). Users are responsible for ensuring that their selected images comply with the applicable licenses and terms of use.

## Configuration

Generation settings are specified in a plant-specific JSON configuration file. An example configuration is shown below:

```json
{
  "num_leaves": 35,
  "diseases": "frogeye, bacterial_blight, rust, cercospora_leaf_blight, downy_mildew, mosaic_virus, potassium_deficiency, sudden_death_syndrome, target_spot, healthy",
  "foliage_size": "(1024, 1500)",
  "single_plant_size": 512,
  "num_plants": 12,
  "disease_rate": 5,
  "background_image_path": "path_to_background_images",
  "input_path": "path_to_input_directory",
  "output_path": "path_to_output_directory",
  "type": "soybean"
}
```

### Configuration fields

| Field | Description |
| --- | --- |
| `num_leaves` | Number of leaves used to construct a plant or foliage sample |
| `diseases` | Comma-separated list of supported disease and health categories |
| `foliage_size` | Dimensions of the generated foliage image |
| `single_plant_size` | Size of each generated plant instance |
| `num_plants` | Number of plants placed in a foliage image |
| `disease_rate` | Disease rate |
| `background_image_path` | Directory containing background images |
| `input_path` | Directory containing transparent individual-leaf images |
| `output_path` | Directory in which generated images will be saved |
| `type` | Plant type; currently, only `soybean` is supported |



## Intended Use

This framework and its generated data are intended for research and educational applications, particularly the development and evaluation of foliar disease classification methods. Performance measured using synthetic images should also be validated on real-world field data before deployment.

## Citation

If you use this framework or its generated dataset, please cite the associated paper:

```bibtex
@article{pakka2026foliagen,
  author={Pakka, Nabin and Wilkerson, Teresa and Chen, Li and Tzeng, Nian-Feng},
  journal={IEEE Access}, 
  title={Foliagen: Framework for Crop Foliage Image Generation from Individual Leaf Images}, 
  year={2026},
  volume={},
  number={},
  pages={1-1},
  keywords={Diseases;Modeling;Industrial plants;Plants (biology);Crops;Training;Accuracy;Computers;Conferences;Image processing;Classifier Training;Crop Leaf Images;Disease Image Classification;Image Generation;Machine Learning},
  doi={10.1109/ACCESS.2026.3727421}}

```