rotate_within_exif fails with image/tiff

[themaikelman]

My actions before raising this issue

• Read/searched the docs
• Searched past issues

Hello,
We are very grateful for this tool, it is really impressive the improvements that are being made in each version, thank you all very much for the effort.

We have updated to version v2.1.0 and we believe that this change #4529 has caused us a bug with tiff images.

Previously the images were uploaded without any problem, however now an exception is raised:

ERROR - 2022-05-24 10:42:53,922 - worker - Traceback (most recent call last):
File "/opt/venv/lib/python3.8/site-packages/rq/worker.py", line 936, in perform_job
rv = job.perform()
File "/opt/venv/lib/python3.8/site-packages/rq/job.py", line 684, in perform
self._result = self._execute()
File "/opt/venv/lib/python3.8/site-packages/rq/job.py", line 690, in _execute
return self.func(*self.args, **self.kwargs)
File "/usr/lib/python3.8/contextlib.py", line 75, in inner
return func(*args, **kwds)
File "/home/django/cvat/apps/engine/task.py", line 662, in _create_thread
preview = extractor.get_preview()
File "/home/django/cvat/apps/engine/media_extractors.py", line 206, in get_preview
return self._get_preview(fp)
File "/home/django/cvat/apps/engine/media_extractors.py", line 130, in _get_preview
preview = rotate_within_exif(preview)
File "/home/django/cvat/apps/engine/media_extractors.py", line 87, in rotate_within_exif
orientation = img.getexif().get(ORIENTATION_EXIF_TAG, ORIENTATION.NORMAL_HORIZONTAL)
File "/opt/venv/lib/python3.8/site-packages/PIL/Image.py", line 1360, in getexif
self._exif.load_from_fp(self.fp, self.tag_v2._offset)
File "/opt/venv/lib/python3.8/site-packages/PIL/Image.py", line 3410, in load_from_fp
self.fp.seek(offset)
AttributeError: 'NoneType' object has no attribute 'seek'
Traceback (most recent call last):
File "/opt/venv/lib/python3.8/site-packages/rq/worker.py", line 936, in perform_job
rv = job.perform()
File "/opt/venv/lib/python3.8/site-packages/rq/job.py", line 684, in perform
self._result = self._execute()
File "/opt/venv/lib/python3.8/site-packages/rq/job.py", line 690, in _execute
return self.func(*self.args, **self.kwargs)
File "/usr/lib/python3.8/contextlib.py", line 75, in inner
return func(*args, **kwds)
File "/home/django/cvat/apps/engine/task.py", line 662, in _create_thread
preview = extractor.get_preview()
File "/home/django/cvat/apps/engine/media_extractors.py", line 206, in get_preview
return self._get_preview(fp)
File "/home/django/cvat/apps/engine/media_extractors.py", line 130, in _get_preview
preview = rotate_within_exif(preview)
File "/home/django/cvat/apps/engine/media_extractors.py", line 87, in rotate_within_exif
orientation = img.getexif().get(ORIENTATION_EXIF_TAG, ORIENTATION.NORMAL_HORIZONTAL)
File "/opt/venv/lib/python3.8/site-packages/PIL/Image.py", line 1360, in getexif
self._exif.load_from_fp(self.fp, self.tag_v2._offset)
File "/opt/venv/lib/python3.8/site-packages/PIL/Image.py", line 3410, in load_from_fp
self.fp.seek(offset)
AttributeError: 'NoneType' object has no attribute 'seek'

Possible Solution

Maybe the image doesn't have EXIF data and that's the problem, or maybe it's related to the PIL function, I'm not sure.

https://github.com/openvinotoolkit/cvat/blob/develop/cvat/apps/engine/media_extractors.py#L86

Steps to Reproduce (for bugs)

1. Try to upload an image/tiff
2. See the error:
   Could not create the task
   Error: Could not create the task on the server. AttributeError: 'NoneType' object has no attribute 'seek'.

[bsekachev]

@atekoa
Hi,
Is it possible to share one image to reproduce the issue?

[themaikelman]

Yes of course, here is a dummy file that reproduces the behavior.
https://we.tl/t-eD7VTvmGOh

[bsekachev]

@atekoa

Looks like there are internal issues with implementation get_exif() in Pillow library, I tried the latest version, but it still persists.

In your case I would suggest to add try except block to rotate_within_exif method:

def rotate_within_exif(img: Image):
    try:
        orientation = img.getexif().get(ORIENTATION_EXIF_TAG,  ORIENTATION.NORMAL_HORIZONTAL)
        if orientation in [ORIENTATION.NORMAL_180_ROTATED, ORIENTATION.MIRROR_VERTICAL]:
            img = img.rotate(180, expand=True)
        elif orientation in [ORIENTATION.NORMAL_270_ROTATED, ORIENTATION.MIRROR_HORIZONTAL_90_ROTATED]:
            img = img.rotate(90, expand=True)
        elif orientation in [ORIENTATION.NORMAL_90_ROTATED, ORIENTATION.MIRROR_HORIZONTAL_270_ROTATED]:
            img = img.rotate(270, expand=True)
        if orientation in [
            ORIENTATION.MIRROR_HORIZONTAL, ORIENTATION.MIRROR_VERTICAL,
            ORIENTATION.MIRROR_HORIZONTAL_270_ROTATED ,ORIENTATION.MIRROR_HORIZONTAL_90_ROTATED,
        ]:
            img = img.transpose(Image.FLIP_LEFT_RIGHT)
    except Exception:
        pass

    return img

And rebuild CVAT adding docker-compose.dev.yml file.

Update:

I found a better approach, if to call img.getexif() before img.thumbnail() (which is called in _get_preview() method), everything works fine. So, in the method _get_preview() (file cvat/apps/engine/media_extractors.py) I swiped two lines

from

        preview.thumbnail(PREVIEW_SIZE)
        preview = rotate_within_exif(preview)

to

        preview = rotate_within_exif(preview)
        preview.thumbnail(PREVIEW_SIZE)

[themaikelman]

Great!! Thanks a lot! I'm trying it right now

[themaikelman]

Yes! It works!! Thank you very much for the help!!