Aluminum Agglomerate Segmentation Dataset

A benchmark dataset for automated detection and characterization of aluminum agglomerates in solid propellant combustion, containing high-speed imagery with instance-level segmentation annotations.

Overview

This dataset accompanies the paper:

Automated Detection and Characterization of Aluminum Agglomerates in Solid Propellant Combustion via Mask R-CNN

Aluminum agglomeration in solid rocket propellants significantly impacts two-phase flow losses and combustion efficiency. This dataset provides manually annotated high-speed images of burning aluminized propellants, enabling the development and benchmarking of automated agglomerate detection algorithms.

Dataset Statistics

Property	Value
Total Images	166
Total Annotations	8,160
Image Resolution	800 × 480 pixels
Image Format	TIFF
Annotation Format	COCO JSON

Pressure Distribution

Pressure	Images	Annotations
2 MPa	86	5,830
3 MPa	39	351
4 MPa	41	1,979

Experimental Setup

Imaging Equipment

• Camera: Chronos 2.1 high-speed camera
• Frame Rate: 5,000–5,400 fps
• Exposure Time: 15 μs
• Spatial Resolution: 10–13 μm/pixel

Combustion Chamber

• Nitrogen-pressurized chamber (~1 L volume)
• 25 mm thick glass window for optical access
• Operating pressure range: 2–4 MPa
• Ignition: Resistive filament (10 W supply)

Propellant Composition

Material	Function	Particle Size	Mass Fraction
HTPB	Fuel binder	—	18%
IPDI	Cross-linker	—	2%
AP (coarse)	Oxidizer	400 μm	52%
AP (fine)	Oxidizer	90 μm	13%
Al (micro)	Metallic fuel	6 μm	10%
Al (nano)	Metallic fuel	50–70 nm	5%

Dataset Structure

Data/
├── images/
│   ├── 2MPa/
│   │   └── frame_XXXXXX.tiff
│   ├── 3MPa/
│   │   └── frame_XXXXXX.tiff
│   └── 4MPa/
│       └── frame_XXXXXX.tiff
├── annotations.json
└── README.md

Annotation Format

Annotations follow the COCO format with instance-level polygonal segmentation masks.

JSON Structure

{
  "images": [
    {
      "id": 186,
      "width": 800,
      "height": 480,
      "file_name": "frame_000165.tiff",
      "path": "/Data/2MPa/frame_000165.tiff",
      "annotated": true,
      "num_annotations": 82,
      "metadata": {
        "pressure": "2MPa",
        "pressure_mpa": 2
      }
    }
  ],
  "annotations": [
    {
      "id": 4545,
      "image_id": 186,
      "category_id": 2,
      "segmentation": [[x1, y1, x2, y2, ...]],
      "area": 32,
      "bbox": [x, y, width, height],
      "iscrowd": false
    }
  ],
  "categories": [
    {
      "id": 2,
      "name": "Particle",
      "supercategory": ""
    }
  ]
}

Annotation Fields

Field	Description
segmentation	List of polygon vertices [x1, y1, x2, y2, ...]
area	Mask area in pixels²
bbox	Bounding box [x_min, y_min, width, height]
iscrowd	Always false (individual instances)

Image Metadata Fields

Field	Description
pressure	Operating pressure ("2MPa", "3MPa", "4MPa")
pressure_mpa	Numeric pressure value (2, 3, or 4)
num_annotations	Count of annotations in this image
annotated	Always true for annotated images

Usage

Loading with PyTorch

import json
from PIL import Image
import numpy as np

# Load annotations
with open('annotations.json', 'r') as f:
    coco_data = json.load(f)

# Create image ID to annotations mapping
img_to_anns = {}
for ann in coco_data['annotations']:
    img_id = ann['image_id']
    if img_id not in img_to_anns:
        img_to_anns[img_id] = []
    img_to_anns[img_id].append(ann)

# Load an image and its annotations
img_info = coco_data['images'][0]
image = Image.open(f"images/{img_info['file_name']}")
annotations = img_to_anns[img_info['id']]

Loading with pycocotools

from pycocotools.coco import COCO

coco = COCO('annotations.json')
img_ids = coco.getImgIds()
ann_ids = coco.getAnnIds(imgIds=img_ids[0])
anns = coco.loadAnns(ann_ids)

Filtering by Pressure

# Filter images by pressure condition
pressure_2mpa = [img for img in coco_data['images'] 
                 if img['metadata']['pressure'] == '2MPa']
pressure_3mpa = [img for img in coco_data['images'] 
                 if img['metadata']['pressure'] == '3MPa']
pressure_4mpa = [img for img in coco_data['images'] 
                 if img['metadata']['pressure'] == '4MPa']

print(f"2 MPa: {len(pressure_2mpa)} images")  # 86 images
print(f"3 MPa: {len(pressure_3mpa)} images")  # 39 images
print(f"4 MPa: {len(pressure_4mpa)} images")  # 41 images

Contact

For questions about the dataset, please open an issue in this repository.