AgriGen is an integrated AI-based engineering system specialized in text-to-image generation for the agricultural domain. The project aims to overcome the limitations of closed-source commercial models by providing an open-source and interpretable solution. The system utilizes deep learning techniques to generate high-quality images of fruits, vegetables, nuts, and seeds from simple textual descriptions.
-
Text-to-Image Generation Generate realistic agricultural crop images from text prompts.
-
Latent Space Exploration Manipulate latent representations to control image characteristics.
-
Dynamic Styling Apply multiple visual styles such as Sketch, Lighting Enhancement, and Grayscale.
-
Interactive Web Interface The system features a user-friendly interface built with Streamlit, allowing users to customize styles and generate images in real-time.
[AgriGen Web Interface]
Figure 1: Screenshot of the AgriGen interactive web application.
The project evolved from an initial VAE-based design into a more advanced implementation for higher image quality.
- Base Model: Stable Diffusion v1.5
- Fine-Tuning Technique: LoRA (Low-Rank Adaptation) for efficient domain-specific training.
- Optimization: Attention Slicing and VAE Slicing for resource-efficient inference.
- Inference Scheduler: DPMSolverMultistepScheduler for faster image generation.
The model was trained on the Fruits-360 Dataset and evaluated using the CLIP model to ensure high semantic alignment.
[Evaluation Samples]
Figure 2: Samples from the evaluation process showing True vs. Predicted labels.
Figure 3: Performance breakdown across different agricultural categories.
Below are sample outputs generated by AgriGen using different prompts and artistic styles.
| Banana | Apple | Tomato |
|---|---|---|
 |
 |
 |
Prompt: "Banana"Style: Natural |
Prompt: "Apple"Style: Lighting Enhancement |
Prompt: "Tomato"Style: Artistic Sketch |