heif-hdrgainmap-decode
Convert HDR photos taken by iPhone 12 (or later) to regular HDR images.
Installation
First, make sure you have the following packages installed:
- Python 3.9 or later
- libheif
- NumPy
- OpenImageIO
On macOS, you can install these packages with Homebrew:
brew install python libheif numpy openimageio
Usage
-
Copy the photo from your iPhone losslessly to your computer.
You can use either AirDrop or a USB cable. But make sure the file is not modified by any software that doesn't support iPhone HDR photo.
Let's assume the filename is
IMG_0000.heic
. -
Extract the HDR gain map from the photo:
heif-convert IMG_0000.heic IMG_0000.png
You will receive two files:
-
IMG_0000.png
: The main image, but we don't need it, because we decode HEIC on our own. -
IMG_0000-urn:com:apple:photo:2020:aux:hdrgainmap.png
: This is the HDR gain map.
-
-
Run the following command to convert the photo to regular HDR image:
./heif-hdrgainmap-decode.py IMG_0000.heic IMG_0000-urn:com:apple:photo:2020:aux:hdrgainmap.png IMG_0000.exr
-
Remove the temporary files.
rm -fv IMG_0000.png IMG_0000-*.png
-
The output is an scRGB (gamma 1.0) encoded OpenEXR file.
You need to use an HDR tone-mapping software to edit it before sharing.
Technical details
The HDR gain map is encoded in the HEIF file as an auxiliary image (auxC
). Both the main image and the HDR gain map is encoded in 8-bits precision. If the image viewer doesn't support this format, it will only show the main SDR image, effectively becoming a tone mapping mechanism.
The exact algorithm is not disclosed, even the Photos app on iPhone and Mac renders differently. Therefore, I assume there is no clear documentation even inside Apple, Inc. The following is a rough guess of how it works:
To render the HDR version of the image, we need to multiply the SDR luminance with “a constant raised to the power of the gain value”. If the gain value is 0.0, the image is unchanged. If the gain value is 1.0 (actually 255), the luminance is multiplied by that constant.
I choose the constant to be 7.5 because it looks good, and it matches scRGB.
License
The program is licensed under the GPLv3 (or later) license. However, my algorithm can be freely used as long as you rewrite the code.
Disclaimer
Please make sure to declare that the algorithm is only a rough guess whenever you distribute or derive this algorithm. This is to prevent misleading anyone that wants to study or improve the algorithm.