Mask-Net

Official PyTorch implementation of
Mask-Net: A U-Net Architecture with Spatial and Channel Attention Mechanisms for Improved Cloud Detection in Sea Surface Temperature Imagery
N'GORAN et al., Artificial Intelligence in Geosciences (under review)

---

Overview

Mask-Net is a U-Net-based deep learning architecture designed for cloud masking in infrared Sea Surface Temperature (SST) imagery. It introduces an Inter-Level Fusion Module (ILFM) that jointly applies spatial and channel attention to encoder skip connections, improving discrimination between clouds and cold oceanic structures such as coastal upwelling fronts — a well-known failure mode of threshold-based cloud detection methods.

Configuration	Resolution	Image size	OA (%)	mIoU (%)	Dice (%)
9 km	[32, 64, 128, 256]	288 × 288	89.1	83.5	86.6
18 km	[32, 64, 128, 256]	144 × 144	90.9	82.8	77.8

---

Architecture

Input SST  (B × 1 × H × W)
      │
 ┌────▼──────────────────────┐
 │  Encoder  (4 stages)      │  →  4 skip connections  s₁…s₄
 │  + Bottleneck             │
 └────┬──────────────────────┘
      │
      ▼
 ┌───────────────────────────┐
 │  SST decoder (ILFM ×4)   │  Joint spatial + channel attention
 │                           │  refines each sᵢ  →  r₁…r₄
 └────┬──────────────────────┘
      │
      ▼
 ┌───────────────────────────┐
 │  Mask decoder (×4)        │  Standard U-Net upsampling using r₁…r₄
 └────┬──────────────────────┘
      │
 Classifier  (1 × 1 conv)
      │
 Logits  (B × 1 × H × W)

Apply torch.sigmoid(logits) > 0.5 to obtain the binary cloud mask.

---

Repository structure

Mask-Net/
├── README.md
├── requirements.txt
├── masknet/
│   ├── __init__.py       # exposes MaskNet, build_masknet
│   ├── model.py          # full architecture
│   ├── ilfm.py           # ILFM + AttentionBlock + _SpatialMLP
│   └── blocks.py         # shared encoder / decoder blocks
├── data.py               # NetCDF loading and train/test split
├── metrics.py            # OA, mIoU, Dice
├── train.py              # training script
└── evaluate.py           # evaluation script

---

Installation

git clone https://github.com/<your-username>/Mask-Net.git
cd Mask-Net
pip install -r requirements.txt

---

Quick start

import torch
from masknet import build_masknet

# 9 km configuration (288 × 288 inputs) — as used in the paper
model = build_masknet(resolution_km=9)

x      = torch.randn(1, 1, 288, 288)
logits = model(x)
mask   = (torch.sigmoid(logits) > 0.5).float()  # binary cloud mask

---

Training

python train.py \
    --image-paths  data/sst_2018.nc \
    --mask-paths   data/gt_2018.nc  \
    --resolution-km 9               \
    --batch-size 8                  \
    --epochs 300                    \
    --checkpoint-dir checkpoints/

Multi-year training (pass one file per year in matching order):

python train.py \
    --image-paths data/sst_2014.nc data/sst_2015.nc ... \
    --mask-paths  data/gt_2014.nc  data/gt_2015.nc  ...

---

Evaluation

python evaluate.py \
    --checkpoint    checkpoints/masknet_9km_final.pth \
    --image-paths   data/sst_2018.nc                  \
    --mask-paths    data/gt_2018.nc

---

Data

The AVHRR Pathfinder SST archive used in the paper is publicly available from NOAA: https://www.ncei.noaa.gov/data/oceans/pathfinder/Version5.3/L3C/

The expert-corrected Ground Truth masks are available from the corresponding author upon reasonable request (see the paper's Data Availability Statement).

---

Citation

If you use this code or the dataset, please cite:

@article{ngoran_masknet_2026,
  title   = {Mask-Net: A U-Net Architecture with Spatial and Channel
             Attention Mechanisms for Improved Cloud Detection in
             Sea Surface Temperature Imagery},
  author  = {N'Goran, Kouassi Adelphe Christian and
             Atiampo, Armand Kodjo and
             Demarcq, Herv{\'e} and
             Cauquil, Pascal and
             Loum, Georges Laussane},
  journal = {Artificial Intelligence in Geosciences},
  year    = {2026},
}

---

Acknowledgements

This work was conducted as part of the PhD thesis of Kouassi Adelphe Christian N'Goran at INPHB (Côte d'Ivoire), in collaboration with IRD and Ifremer (MARBEC, France).
Partially funded by the France Excellence 2025 grant (SCAC, Embassy of France in Abidjan) and by the Direction de l'Orientation et des Bourses (DOB) of the Ivorian Ministry of Higher Education and Scientific Research.

---

License

MIT License — see LICENSE for details.