PUSHPENDAR/hrsid-ship-detection

🚢 HRSID Ship Detection Demo

This Gradio Space demonstrates a Faster R-CNN model trained on HRSID dataset for ship detection in SAR imagery.

Upload a SAR image to detect ships.

---

license: apache-2.0 tags: - object-detection - synthetic-aperture-radar - detectron2 - faster-rcnn - ship-detection datasets: - HRSID metrics: - mAP

🚢 Ship Detection in SAR Imagery using Faster R-CNN (HRSID)

Abstract

Synthetic Aperture Radar (SAR) imagery enables all-weather, day-and-night maritime monitoring. However, ship detection in SAR images is challenging due to speckle noise, varying vessel scales, and complex coastal backgrounds.

This work presents a Faster R-CNN (ResNet-50 FPN) based object detection model trained on the HRSID dataset for robust ship detection in high-resolution SAR imagery.

---

📷 Qualitative Results

Detection Output

The model successfully detects ships of varying scales in cluttered maritime environments.

---

🧠 Model Architecture

• Framework: Detectron2
• Detector: Faster R-CNN
• Backbone: ResNet-50 + Feature Pyramid Network (FPN)
• Classes: 1 (Ship)
• Inference Device: CPU compatible

---

📊 Training Configuration

Parameter	Value
Dataset	HRSID
Image Resolution	1400 × 1400
Iterations	{MAX_ITER}
Score Threshold	{SCORE_THRESH}
Optimizer	SGD
Learning Rate	0.0025

---

📈 Evaluation

Metric	Value
mAP@0.5	(Add your value)
Precision	(Add value)
Recall	(Add value)

Note: Metrics computed on validation split of HRSID dataset.

---

🛰 Dataset Description

HRSID (High-Resolution SAR Images Dataset) contains:

• 5000+ SAR images
• Thousands of annotated ship bounding boxes
• Multi-scale vessel distribution
• Coastal and open-sea scenarios

---

⚙ Inference Code

from detectron2.config import get_cfg
from detectron2.engine import DefaultPredictor
import cv2

cfg = get_cfg()
cfg.merge_from_file("config.yaml")
cfg.MODEL.WEIGHTS = "model_final.pth"
cfg.MODEL.ROI_HEADS.SCORE_THRESH_TEST = 0.5

predictor = DefaultPredictor(cfg)

image = cv2.imread("test_sar.jpg")
outputs = predictor(image)
print(outputs)