FR-DEEP Batched Dataset
The FR-DEEP Batched Dataset is a dataset of labeled radio galaxies suitable for use with deep learning algorithms. The labels for the samples are compiled from the FRICAT and CoNFIG catalogs. Each sample is classified as either Fanaroff-Riley Class I (FRI) or Class II (FRII). This dataset forms the base training data for the paper Transfer Learning for Radio Galaxy Classification. If you use this dataset please cite:
(1) Transfer learning for radio galaxy classification, Tang H., Scaife A. M. M., Leahy J. P., 2019, arXiv:1903.11921
The FR-DEEP Batched Dataset is comprised of two separate sub-datasets: FRDEEP-N and FRDEEP-F. The two subsets provide images of the same objects taken from (1) the NVSS survey and (2) the FIRST survey. Each subset contains 600 150x150 images in two classes: FR I & FR II. Images were extracted from the Skyview Virtual Observatory, and underwent pre-processing described in (1).
The angular size of the pixels for each dataset is: FRDEEP-N: 15’‘/pixel; FRDEEP-F: 1.8’‘/pixel. In terms of angular scale, this means that a 150 x 150 pixel FIRST image covers the same area as an 18 x 18 pixel NVSS image.
There are 550 training images, and 50 test images. The FR-DEEP dataset is inspired by CIFAR-10 Dataset and HTRU1 Dataset.
The dataset is divided into 11 training batches and 1 test batch. Each batch contains 50 images. In total the dataset contains 264 FR I objects and 336 FR II objects. In each batch there are 22 FR I and 28 FR II images, organized in random order.
This is an imbalanced dataset
NVSS set images look like:
FR I:
FR II:
FIRST set images, on the other hand, look like:
FR I:
FR II:
The FRDEEP.py file contains an instance of the torchvision Dataset() for the FRDEEP Batched Dataset.
To use it with PyTorch in Python, first import the torchvision datasets and transforms libraries:
from torchvision import datasets
import torchvision.transforms as transforms
Then import the FRDEEPN class:
from FRDEEP import FRDEEPN
Define the transform:
# convert data to a normalized torch.FloatTensor
transform = transforms.Compose(
[transforms.ToTensor(),
transforms.Normalize([0.5],[0.5])])
Read the FRDEEPN dataset from ‘/NVSS_data’:
# choose the training and test datasets
train_data = FRDEEPN('/NVSS_data', train=True, download=False, transform=transform)
test_data = FRDEEPN('/NVSS_data', train=False, download=False, transform=transform)
An example of classification using the FRDEEPN class in PyTorch is provided as a Jupyter notebook using the NVSS dataset or using the FIRST dataset.
These are examples for demonstration only - please don’t use them for science!