Extracting EXIF data increases memory consumption lead finally to a silent die of Pythonista

[shield61]

Hi, first of all, I am very impressed by Pythonista and the tons of possibilities it offers for me on my iOS devices and I am as well impressed from the community around. I am really looking forward for the next release.

As an amateur photographer I liked the possibilities an iPad has to offer. It is easy to integrate external cameras and is a lightweight solution on travel. Apple shortcuts, Pythonista, iSH and a-shell are a magic bag for automation.

Travelling within different timezones and sometimes forgot to collect GPS data for my external camera, I experienced "inconsistencies" in the metadata of the images in my iOS Photolibrary.
This is what I want to correct but have to analyze first. Therfore I need the EXIF data from the images of my trip in my Photolibrary including imported images as well as from HEIC image taken by my iPhone.

I believe I have researched nearly all sources I found to extract metadata. Best method so far I found is described in this article:
https://www.gfrigerio.com/read-exif-data-of-pictures/. Combined with some code from @omz written in photopicker2.py and hints and code from @cvpe I was able to get all metadata of images in the Photolibrary including HEIC images.
Basically this is done with help of CIImage:

    def exif(self):
        """Return EXIF data of the Asset if existant"""
        info = self.fetch_data()
        img_data = info['data']
        if self._exif is not None:
            return self._exif
        # Extract EXIF data from Asset via CIImage
        cImage = None
        CIImage = ObjCClass('CIImage')
        cImage = CIImage.imageWithData_(img_data) if img_data else None
        if not cImage:
            return None
        props = None
        props = cImage.properties()
        if props is None:
            result = None
        else:
            self._exif = parse_ns_dictionary_string(str(props['{Exif}']))
        props = None  # Release object
        cImage = None  # Release object
        if cImage: del cImage  # Release object
        if props: del props
        if hasattr(self, '_exif'):
            return self._exif
        return None

I call the exif method for every asset in the Photolibrary for a day of my trip and can repeat it for more days.
On my Macbook I can see in the ActivityMonitor memory consumption increases fast (depending on the number of images iterated). One of the next iterations Pythonista dies silently. Same on iPad, but I cannot see the memory consumption.
I hope I have released all objects in Python and I know I do not have any possibility to directly release iOS objects.

But may be someone here in the community has an idea how to overcome the memory increase.
Any help and hints are really appreciated.
(The whole demonstration code can be found in a Gist: https://gist.github.com/shield61/93deb01c979d2ffe83f75cf5efc2b1fc)
Thanks in advance.

[qqfunc]

I'm not sure if this will help you, but I recommend using Rubicon-ObjC instead of objc_util. objc_util is based on an outdated version of Rubicon-ObjC, which is now much more stable and convenient. I believe it also resolves some memory-related issues such as beeware/rubicon-objc#256.

There are several ways to use Rubicon-ObjC in Pythonista:

• Install from PyPI using a tool like StaSH or Pythonista3 pip configuration tool.
• Manually download the package and move to the site-packages directory.
• Use the Pythonista Lab app (beta).

[RichardPotthoff]

Here is a description of a method I have used to install "pip" in Pythonista 3.4:
Pythonista 3.4: 'pip install' (alternative to StaSh)

[shield61]

@RichardPotthoff Thanks I will definitely try this. Has you successfully installed and used rubicon.objc with that method?

[RichardPotthoff]

Yes, I was able to install rubicon-objc, and import rubicon.objc, but I got an error message like:"Calendar Class Instance not callable", when I tried to use it. May be there are differences between objc_util and rubicon.objc.

[shield61]

Ok thanks.

[RichardPotthoff]

Here is the error I get in Pythonista Lab(beta) with objc_util replaced with rubicon.objc:
(Pythonista Lab(beta) has rubicon-objc pre-installed)

---> 86 calendar = NSCalendar.currentCalendar()
     87 return calendar.dateFromComponents_(comps)

TypeError: 'ObjCInstance' object is not callable

This is the same error I got in Pythonista3.4 when using rubicon.objc.

To me it looks like what used to be methods in objc_util are now attributes or properties in rubicon.objc.

[mmccthomas]

I wonder if the autorelease method might be the key to this?
E.g. VNImageRequestHandler = ObjCClass('VNImageRequestHandler')
handler = VNImageRequestHandler.alloc().initWithData_options_(img_data, None).autorelease()

I’m no expert on Objective C integration, but this code has helped me several times to highlight the locations of memory leaks
Import resource
resource.getrusage(resource.RUSAGE_SELF).ru_maxrss // (2 ** 20)

Hope either of these might help
Best Regards
Chris Thomas

[shield61]

Thanks for the reminder to ask ChatGPT for this issue. I used it a lot, but not exactly for this problem. In fact I asked now and the result is promising.
👉 Recommendation: Since you’re extracting Exif only, the CGImageSourceCopyPropertiesAtIndex approach is the standard and most memory-efficient.
ChatGPT generated some Python code:

from objc_util import ObjCClass, nsdata

# CoreGraphics / ImageIO classes
CGImageSourceCreateWithData = ObjCClass('CGImageSource').createWithData_options_
CGImageSourceCopyPropertiesAtIndex = ObjCClass('CGImageSource').copyPropertiesAtIndex_options_

def get_metadata(image_bytes):
    ns_data = nsdata(image_bytes)
    source = CGImageSourceCreateWithData(ns_data, None)
    if source:
        props = CGImageSourceCopyPropertiesAtIndex(source, 0, None)
        return dict(props)
    return {}

I will try this tomorrow, after I am back home and give feedback.

[shield61]

@mmccthomas you mean commenting out

• e = threading.Event()
• e.set()
• e.wait()
  I have different line numbers, but I commented them out, unfortunately Pythonista dies after some iterations.
  But this was a quick try, I will do a more intense try

[shield61]

@mmccthomas I appreciate that you stick to the issue.

[shield61]

@mmccthomas I checked teh resource consumption with 'resource.getrusage'. It shows continuously increasing values.

[mmccthomas]

You’re right. On my system it alternates on each run between increasing memory and static 1GB memory. In the latter case it crashes after a while. Sorry, I don’t have further insights. It seems to be a buffer with in the CIImage object. Above my pay grade!

[cvpe]

Hello, almost sure this post would not help you, but in the past I have written a lot of code in Pythonista for my photos, and to get their exifs, I always had used piexif.py, see https://github.com/hMatoba/Piexif. Perhaps you could have a try.

[shield61]

Piexif was also my first choice, but does not work on HEIC images.

[mmccthomas]

I don’t suppose that https://github.com/exif-heic-js/exif-heic-js could be converted to Python?
There is also a package ExifRead that claims to read HEIC
I tried this and it appeared to work :

import exifread 

im = Image.open(asset.get_image_data())
im.show()
tags = exifread.process_file(asset.get_image_data())
for tag, value in tags.items():
if tag not in ('JPEGThumbnail', 'TIFFThumbnail', 'Filename', 'EXIF MakerNote'):
        print(f"Key: {tag}, value {value}")

Key: Image Orientation, value Horizontal (normal)
Key: Image XResolution, value 72
Key: Image YResolution, value 72
Key: Image ResolutionUnit, value Pixels/Inch
Key: Image YCbCrPositioning, value Centered
Key: Image ExifOffset, value 102
Key: Thumbnail Compression, value JPEG (old-style)
Key: Thumbnail XResolution, value 72
Key: Thumbnail YResolution, value 72
Key: Thumbnail ResolutionUnit, value Pixels/Inch
Key: Thumbnail JPEGInterchangeFormat, value 286
Key: Thumbnail JPEGInterchangeFormatLength, value 5991
Key: EXIF ExifVersion, value 0221
Key: EXIF ComponentsConfiguration, value YCbCr
Key: EXIF FlashPixVersion, value 0100
Key: EXIF ColorSpace, value sRGB
Key: EXIF ExifImageWidth, value 5712
Key: EXIF ExifImageLength, value 4284
Key: EXIF SceneCaptureType, value Standard    

[shield61]

HI @mmccthomas I am pretty sure I have tried this too and found the EXIF metadata for HEIC is quite minimalistic. But only if you take the image data form asset.get_image_data. If you open a HEIC file it is much better.

By now I have tried the solution suggested by ChatGPT, with some code generated.
From the tests I have made so far I can say the memory consumption (not the value I get from 'resource.getrusage') is flat. I could go in the loop far beyond I ever was.
I have updated my gist with the updated code (https://gist.github.com/shield61/93deb01c979d2ffe83f75cf5efc2b1fc).

I will try afterwords the ExifRead method again. May be my memory was not correct and the continue with using rubicon.objc.

Thanks to all for your help, for now at least I have one working solution and still a lot of input to try.

[mmccthomas]

So glad that you have a workaround.
Best Regards,
Chris