How abdul-81-basha/oceanmind-ai works
OceanMind is a marine-pollution segmentation platform: users upload an 11-band optical GeoTIFF, the API runs a 15-class U-Net prediction, and the frontend presents maps, class breakdowns, and exportable results. The repository also contains the model’s dataset, training, and evaluation workflow. The frontend and backend contract is documented and excerpted; training and evaluation relationships are supported by their source-index references. The frontend excerpt is partial, so its internal UI implementation is represented as one component.
Main entry points: main.py Prediction API · train_model.py Model Training · evaluate_model.py Model Evaluation · page.tsx Analysis Console
Components 9
User Experience
- Analysis Console
app/page.tsx
Inference Service
- Prediction API
backend/main.py - U-Net Model
unet_model.py - Prediction Assets
backend/main.py
Model Development
- Model Checkpoint
outputs/marine_debris_segmentation.pth - MARIDA Dataset Loader
dataset_loader.py - Model Training
train_model.py - Model Evaluation
evaluate_model.py
Other components
- Analyst
Connections 12
- Analyst → Analysis Console: uploads imagery
README.md - Analysis Console → Prediction API: posts image
app/page.tsx - Prediction API → U-Net Model: runs inference
backend/main.py - Prediction API → Prediction Assets: writes maps
backend/main.py - Analysis Console → Prediction Assets: loads result images
README.md - Model Training → MARIDA Dataset Loader: loads samples
train_model.py - Model Training → U-Net Model: trains model
train_model.py - Model Training → Model Checkpoint: saves weights
train_model.py - Model Evaluation → MARIDA Dataset Loader: loads test data
evaluate_model.py - Model Evaluation → U-Net Model: evaluates model
evaluate_model.py - Model Evaluation → Model Checkpoint: loads weights
evaluate_model.py - Prediction API → Model Checkpoint: loads weights
backend/main.py