A comprehensive image dataset for fish and shrimp disease classification containing 5,887 high-quality images across 11 disease categories. This dataset is designed to support machine learning models for early detection of common aquaculture diseases in Bangladesh.

Dataset Overview

Feature	Description
Task	Image Classification
Size	5,887 images
Classes	11 disease conditions
Format	JPG/PNG images
Resolution	Various (original captures)
License	CC BY-NC-SA 4.0
Domain	Aquaculture, Fish Pathology

Dataset Structure

The dataset is split into standard train/validation/test sets:

Split	Images	Percentage
Train	4,118	70%
Validation	1,175	20%
Test	594	10%

Disease Classes

This dataset contains the following disease categories:

Fish Diseases (7 classes)

Fish_Bacterial Red disease - 297 images
A bacterial infection causing red lesions on the fish body
Fish_Bacterial diseases - Aeromoniasis - 296 images
Caused by Aeromonas bacteria, common in freshwater fish
Fish_Bacterial gill disease - 290 images
Infection affecting gill tissue, causing respiratory distress
Fish_Fungal diseases Saprolegniasis - 293 images
Fungal infection appearing as cotton-like growths
Fish_Healthy Fish - 300 images
Normal, disease-free fish for comparison
Fish_Parasitic diseases - 303 images
Various parasitic infections common in aquaculture
Fish_Viral diseases White tail disease - 303 images
Viral infection characterized by whitening of the tail area

Shrimp Diseases (4 classes)

Shrimp_Black_Gill - 555 images
Condition where shrimp gills turn black due to various factors
Shrimp_Healthy - 2120 images
Normal, disease-free shrimp for comparison
Shrimp_White_Spot_Syndrome_Virus - 545 images
Highly contagious viral disease causing white spots
Shrimp_White_Spot_Syndrome_Virus_and_Black_Gill - 585 images
Co-infection of both WSSV and black gill condition

Class Distribution

Training Set:
  Fish_Bacterial Red disease: 207 images
  Fish_Bacterial diseases - Aeromoniasis: 207 images
  Fish_Bacterial gill disease: 203 images
  Fish_Fungal diseases Saprolegniasis: 205 images
  Fish_Healthy Fish: 210 images
  Fish_Parasitic diseases: 212 images
  Fish_Viral diseases White tail disease: 212 images
  Shrimp_Black_Gill: 388 images
  Shrimp_Healthy: 1484 images
  Shrimp_White_Spot_Syndrome_Virus: 381 images
  Shrimp_White_Spot_Syndrome_Virus_and_Black_Gill: 409 images

Validation Set:
  Fish_Bacterial Red disease: 59 images
  Fish_Bacterial diseases - Aeromoniasis: 59 images
  Fish_Bacterial gill disease: 58 images
  Fish_Fungal diseases Saprolegniasis: 58 images
  Fish_Healthy Fish: 60 images
  Fish_Parasitic diseases: 60 images
  Fish_Viral diseases White tail disease: 60 images
  Shrimp_Black_Gill: 111 images
  Shrimp_Healthy: 424 images
  Shrimp_White_Spot_Syndrome_Virus: 109 images
  Shrimp_White_Spot_Syndrome_Virus_and_Black_Gill: 117 images

Test Set:
  Fish_Bacterial Red disease: 31 images
  Fish_Bacterial diseases - Aeromoniasis: 30 images
  Fish_Bacterial gill disease: 29 images
  Fish_Fungal diseases Saprolegniasis: 30 images
  Fish_Healthy Fish: 30 images
  Fish_Parasitic diseases: 31 images
  Fish_Viral diseases White tail disease: 31 images
  Shrimp_Black_Gill: 56 images
  Shrimp_Healthy: 212 images
  Shrimp_White_Spot_Syndrome_Virus: 55 images
  Shrimp_White_Spot_Syndrome_Virus_and_Black_Gill: 59 images

Usage

Loading the Dataset

from datasets import load_dataset

# Load the entire dataset
dataset = load_dataset("Saon110/bd-fish-disease-dataset")

# Access specific splits
train_dataset = dataset["train"]
val_dataset = dataset["valid"]
test_dataset = dataset["test"]

# Get an image and its label
image = train_dataset[0]["image"]
label = train_dataset[0]["label"]
label_name = train_dataset[0]["label_name"]

# Display an example image
import matplotlib.pyplot as plt
plt.imshow(image)
plt.title(f"Class: {label_name} (Label ID: {label})")
plt.axis("off")
plt.show()

Training a Simple Classifier

import torch
from torch.utils.data import DataLoader
from torchvision import transforms
from transformers import ViTForImageClassification, ViTImageProcessor

# Load the dataset
dataset = load_dataset("Saon110/bd-fish-disease-dataset")

# Define image transformations
image_processor = ViTImageProcessor.from_pretrained("google/vit-base-patch16-224")

def transform_images(examples):
    examples["pixel_values"] = image_processor(examples["image"], return_tensors="pt")["pixel_values"]
    return examples

# Apply transformations
dataset = dataset.map(transform_images, batched=True)

# Format dataset for training
dataset = dataset.with_format("torch", columns=["pixel_values", "label"])

# Create data loaders
train_dataloader = DataLoader(dataset["train"], batch_size=16, shuffle=True)
val_dataloader = DataLoader(dataset["valid"], batch_size=16)

# Load pretrained model
model = ViTForImageClassification.from_pretrained(
    "google/vit-base-patch16-224", 
    num_labels=11,
    id2label={str(i): label for i, label in enumerate(dataset["train"].features["label_name"].names)},
    label2id={label: str(i) for i, label in enumerate(dataset["train"].features["label_name"].names)}
)

# Continue with your training loop...

Dataset Creation

This dataset is a curated combination of two high-quality aquaculture disease datasets:

Source Datasets

Shrimp Disease Dataset: Mendeley Data Repository
- Contains shrimp disease images including Black Gill, White Spot Syndrome Virus, and healthy specimens
- Original research data from controlled aquaculture studies
Fish Disease Dataset: HuggingFace - panda992/fish_disease_datasets
- Comprehensive collection of fish disease images covering bacterial, viral, fungal, and parasitic conditions
- Includes healthy fish specimens for comparison

Dataset Preparation Process

The creation of this unified dataset involved:

Data Integration: Combining fish and shrimp disease images from the two source datasets
Quality Control: Filtering and validating image quality and label accuracy
Standardization: Consistent naming conventions and class labeling across both domains
Expert Review: Verification of disease classifications by aquaculture specialists
Train/Validation/Test Splitting: Professional 70%/20%/10% split maintaining class balance

Importance and Applications

This dataset addresses a critical need in Bangladesh's aquaculture industry, where early disease detection can:

Reduce economic losses (estimated $100M+ annually)
Minimize antibiotic usage through targeted treatment
Improve food security in a country where fish provides 60% of animal protein
Enable automated monitoring solutions for small-scale farmers

Citation

If you use this dataset in your research, please cite:

@dataset{saon110_bd_fish_disease_2025,
  author = {Sijon Chisty Saon},
  title = {BD Fish & Shrimp Disease Dataset},
  year = {2025},
  publisher = {HuggingFace},
  url = {https://huggingface.co/datasets/Saon110/bd-fish-disease-dataset}
}

License

This dataset is released under CC BY-NC-SA 4.0 license, allowing non-commercial use with attribution. For any commercial use, contact me : saonchishty@gmail.com

Acknowledgments

We thank the original dataset creators for making their research data publicly available:

Shrimp Disease Dataset: Thanks to the researchers who contributed the Mendeley dataset on shrimp diseases (jhrtdj9txm/1)
Fish Disease Dataset: Appreciation to panda992 for the comprehensive fish disease collection on HuggingFace

Special thanks to:

The Department of Fisheries, Bangladesh
Participating fish farmers and aquaculture facilities
Aquaculture disease specialists who assisted with disease verification and validation
The open science community for promoting data sharing in agricultural research

Downloads last month: 71

Models trained or fine-tuned on Saon110/bd-fish-disease-dataset

Saon110/fish-shrimp-disease-classifier

Image Classification • 85.8M • Updated Sep 19, 2025 • 19