DETR-BASE_Marine

Overview

• Model Name: DETR-BASE_Marine
• Model Architecture: DETR (End-to-End Object Detection) with ResNet-50 backbone.
• Model Type: Object Detection
• Framework: PyTorch
• Dataset: Aerial Maritime Image Dataset
• License: MIT License (for the dataset)

Model Description

The DETR-BASE_Marine Aerial Maritime Detector is a deep learning model based on the DETR architecture with a ResNet-50 backbone. It has been fine-tuned on the "Aerial Maritime Image Dataset," which comprises 74 aerial photographs captured via a Mavic Air 2 drone. The model is designed for object detection tasks in maritime environments and can identify and locate various objects such as docks, boats, lifts, jetskis, and cars in aerial images.

Key Features:

• Multi-class object detection.
• Object classes: Docks, Boats, Lifts, Jetskis, Cars.
• Robust performance in aerial and maritime scenarios.

Use Cases

• Boat Counting: Count the number of boats on water bodies, such as lakes, using drone imagery.
• Boat Lift Detection: Identify the presence of boat lifts on the waterfront via aerial surveillance.
• Car Detection: Detect and locate cars within maritime regions using UAV drones.
• Habitability Assessment: Determine the level of inhabitation around lakes and water bodies based on detected objects.
• Property Monitoring: Identify if visitors or activities are present at lake houses or properties using drone surveillance.
• Proof of Concept: Showcase the potential of UAV imagery for maritime projects and object detection tasks.

Dataset

• Dataset Name: Aerial Maritime Image Dataset
• Number of Images: 74
• Number of Bounding Boxes: 1,151
• Collection Method: Captured via Mavic Air 2 drone at 400 ft altitude.

Usage

from transformers import DetrImageProcessor, DetrForObjectDetection
import torch
from PIL import Image

img_path = ""
image = Image.open(img_path)

extractor = AutoFeatureExtractor.from_pretrained("TuningAI/DETR-BASE_Marine")
model = AutoModelForObjectDetection.from_pretrained("TuningAI/DETR-BASE_Marine")

inputs = processor(images=image, return_tensors="pt")
outputs = model(**inputs)

# convert outputs (bounding boxes and class logits) to COCO API
# let's only keep detections with score > 0.9
target_sizes = torch.tensor([image.size[::-1]])
results = processor.post_process_object_detection(outputs, target_sizes=target_sizes, threshold=0.9)[0]

for score, label, box in zip(results["scores"], results["labels"], results["boxes"]):
    box = [round(i, 2) for i in box.tolist()]
    print(
            f"Detected {model.config.id2label[label.item()]} with confidence "
            f"{round(score.item(), 3)} at location {box}"
    )

License

This model is provided under the MIT License.

The Aerial Maritime Image Dataset used for fine-tuning is also under the MIT License.