Tutorial - How To Use

Tensorflow 2.x FITS was initiall designed to work like Tensorflow’s built in image decoders. If you are used to using those it should hopefully feel familiar. Read in the fits file as a binary string using Tensorflow’s tf.io.read_file and then decode with the relavent tf_fits decoder. All decoders take the binary string as the first argument and the header number (zero indexed) as the second argument:

XXX_decode_fits(binary_string, header)

where XXX is image, bintable or asciitable (see below).

These functions return an N-dimensional tf.float32 tensor.

Currently, the header type (image, binary table or ascii table) is not checked, so make sure your header value is correct!

FITS Images

The returned image shape will be the same as the image in the FITS (i.e. 2D if the FITS HDU is 2D or 3D if it is 3D and so on). It will be returned as a tf.float32 tensor so there may be data loss if the image is stored as double precision floats.

FITS images can be loaded as demonstrated:

import tensorflow as tf
from tf_fits.image import image_decode_fits

fits_file = '/path/to/fits/file.fits'
header = 0

img = tf.io.read_file(fits_file)
img = image_decode_fits(img, header)

FITS Binary Tables

The returned binary table will have the same number of rows and columns as the binary table in the FITS. HOWEVER, data types will not always be conserved. Character strings (‘A’), complex numbers (‘C’ and ‘M’), array descriptors (‘P’ and ‘Q’) and bits (‘X’) will be converted into single-precision floating point values due to the way Tensorflow seems to want to work. Arrays of any data type (e.g. ‘5E’) will be returned as the first value in the array. For example, columns of data type ‘5E’ will become columns of data type ‘E’ by taking the first value in the array. Non-single-precision floating point values will also be converted to single-precision floating points.

FITS binary tables can be loaded as demonstrated:

import tensorflow as tf
from tf_fits.bintable import bintable_decode_fits

fits_file = '/path/to/fits/file.fits'
header = 1

tbl = tf.io.read_file(fits_file)
tbl = bintable_decode_fits(tbl, header)

FITS ASCII Tables

The returned ascii table will have the same number of rows and columns as the ascii table in the FITS. HOWEVER, data types will not always be conserved. Character strings (‘A’) will be converted into single-precision floating point values due to the way Tensorflow seems to want to work. Non-single-precision floating point values will also be converted to single-precision floating points.

FITS ASCII tables can be loaded as demonstrated:

import tensorflow as tf
from tf_fits.asciitable import asciitable_decode_fits

fits_file = '/path/to/fits/file.fits'
header = 1

tbl = tf.io.read_file(fits_file)
tbl = asciitable_decode_fits(tbl, header)