Sandesh-Lav/potato-caption-blip

---

Potato Tuber Disease Captioning Model (BLIP)

This model is a fine-tuned version of Salesforce/blip-image-captioning-base trained to generate descriptive captions for potato tuber diseases.

The model analyzes an input image of a potato tuber and produces a natural language description of the visible disease symptoms.

Example output:

A cut-section of a potato tuber showing moderate dry rot within the internal tissue.

---

Dataset

Training was performed using the Sandesh-Lav/potato-tuber-caption-dataset.

Dataset characteristics:

Property	Value
Total images	~1846
Train split	1661
Validation split	92
Test split	93

Each image is paired with a scientific description of potato tuber diseases, including:

• bacterial soft rot
• dry rot
• greening disorder
• internal tissue damage

---

Training

The model was fine-tuned using the Hugging Face Transformers Trainer.

Training configuration:

Parameter	Value
Base model	BLIP image captioning
Epochs	5
Batch size	4
Learning rate	5e-5
Optimizer	AdamW

Training loss after fine-tuning:

Training loss ≈ 0.19

---

Evaluation

Evaluation was performed on the held-out test set.

Metrics used:

Metric	Score
BLEU-4	0.74
ROUGE-L	0.95
METEOR	0.92
CIDEr	5.43

The high scores indicate that the generated captions closely match the reference disease descriptions.

Note: The dataset contains structured scientific captions, which may lead to higher similarity scores compared to open-domain caption datasets.

---

Usage

Load the model using Hugging Face Transformers:

from transformers import BlipProcessor, BlipForConditionalGeneration
from PIL import Image

processor = BlipProcessor.from_pretrained("Sandesh-Lav/potato-caption-blip")
model = BlipForConditionalGeneration.from_pretrained("Sandesh-Lav/potato-caption-blip")

image = Image.open("potato.jpg").convert("RGB")

inputs = processor(images=image, return_tensors="pt")

out = model.generate(**inputs, max_new_tokens=40)

caption = processor.decode(out[0], skip_special_tokens=True)

print(caption)

Example output:

A potato tuber showing severe bacterial soft rot with tissue degradation.

---

Applications

This model can be used for:

• agricultural disease documentation
• automated crop disease reporting
• smart farming systems
• agricultural AI research

---

Limitations

• The dataset contains structured captions describing potato tuber diseases.
• The model may struggle with images outside this domain.
• Performance may decrease on different lighting conditions or potato varieties.

---

Citation

If you use this model, please cite the dataset and BLIP architecture.

---

Author

Developed by as part of an academic mini-project.

---