Advanced Fisheries Monitoring

State-of-the-art computer vision to detect and classify fish species from boat footage. Powered by ResNet50 and Vision Transformers.

Get Started View on GitHub

Dual Architecture Power

🧠

CNN (ResNet50)

Robust feature extraction using a pre-trained ResNet50 backbone. Optimized for simultaneous classification (8 classes) and bounding box regression.

👁️

Transformer (ViT)

Leveraging self-attention mechanisms with `vit_base_patch16_224` for superior global context understanding and localization.

Gradio Interface

Interactive web application for real-time inference. Upload images and instantly get species classification and bounding boxes.

Getting Started

Set up your environment in seconds.

Installation
conda create -n fisheries python=3.10
conda activate fisheries
pip install -r requirements.txt

Train the models using our optimized scripts.

Train CNN
python src/train.py --model_type cnn --epochs 10 --learning_rate 1e-4
Train ViT
python src/train.py --model_type vit --epochs 10 --learning_rate 5e-5