I've changed the type casting of tags as strings decoded using an encoding parameter passed to pytiff.Tiff. This will prevent an error on unicode tags in python 3 and will resolve issue #16. This new behavior will allow users to decode the tag values into their preferred encoding. Setting tags also uses the encoding parameter.

I found out after installing pytiff that the package cython was missing as a dependency because I did not install it previously.

The error message (under Windows and Python 3.7.2) with pytiff-0.8.0 said "Cython modules not available". Could you please make sure that during the installation the installer setup checks the availabillity of cython?

I encountered this proplem as I tried to save a 3D np.array as a stack of tifs.

What I essentially did was this:

img_stack = np.zeros((100,100,50), dtype = 'uint8')
img_stack[30:60,30:60,20:30] = 255

for ind in range(50):
    img_slice = img_stack[:,:,ind]
    with pytiff.Tiff(file_path.format(ind), 'w') as handle:
        handle.write(img_slice, method='scanline')

This produces same strange artefacts, when used on a regular image stack. On this particular 'image stack' it should produce all black tifs, if the problem is reproduceable on other machines.

The problem does not occur, when the flag method is set to 'tile'. I fixed this py replacing the line img_slice = img_stack[:,:,ind] with img_slice = img_stack[:,:,ind].copy()

pip install pytiff
Downloading/unpacking pytiff Downloading pytiff-0.7.0.tar.gz (1.3MB): 1.3MB downloaded Running setup.py (path:/tmp/pip-build-3abmxk7b/pytiff/setup.py) egg_info for package pytiff Traceback (most recent call last): File "/tmp/pip-build-3abmxk7b/pytiff/setup.py", line 13, in import numpy ImportError: No module named 'numpy'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "<string>", line 17, in <module>
  File "/tmp/pip-build-3abmxk7b/pytiff/setup.py", line 15, in <module>
    raise Exception("Numpy is needed for installation")
Exception: Numpy is needed for installation
Complete output from command python setup.py egg_info:
Traceback (most recent call last):

File "/tmp/pip-build-3abmxk7b/pytiff/setup.py", line 13, in

import numpy

ImportError: No module named 'numpy'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):

File "", line 17, in

File "/tmp/pip-build-3abmxk7b/pytiff/setup.py", line 15, in

raise Exception("Numpy is needed for installation")

Exception: Numpy is needed for installation

I was looking for an example of a 'tiled' Bigtiff file, so naturally I opened test_data/bigtif_example_tiled.tif from this repo:

In [90]: hex( ord( open('test_data/bigtif_example_tiled.tif', 'r').read(4)[2] ) )
Out[90]: '0x2a'

Whoops, the third byte of a Bigtiff file should be 0x2b, not 0x2a.[1]

[1]: http://bigtiff.org/

I noticed that pytiff spends a lot of time for padding data before writing it to file. In the current implementation, padding is always applied, even if the padding is zero in all dimensions.

I modified the implementation to only pad when padding is actually necessary. In the process, I noticed that pytiff seems to "abuse" the padding to convert the incoming buffer from Fortran memory layout to C memory layout. Thus, if no padding is applied, we have to do this conversion explicitly.

The check for padding is not the prettiest, but using np.sum instead introduces a significant overhead.

In my tests (writing a chunked TIFF file, 6000x7000 pixels, plus pyramid), the change reduces the time for pytiff's __setitem__ from 8.93 seconds to 0.807 seconds (so ~10 fold speedup). The results are identical (verified by checksum).

Pytiff gives an OSError: file not found when opening a new image with pytiff.Tiff('filename.tif') in case that the filename path is a symlink. I could overcome this problem by doing from os import readlink; pytiff.Tiff(readlink('filename.tif')); however this is not the expected behaviour.

I am trying to read a large Aperio Imagescope image (BigTIFF format). I can read the image without issues and it is very fast. However, after I am finished, I am running into an error that

 Fatal Python error: deallocating NotImplemented

Has this issue ever come up before? Any way to avoid it? Thanks

Although TIFF string tags are only supposed to contain 7-bit ASCII characters, many tools write values in UTF-8 or other encodings that aren't 7-bit clean. OME-TIFF/BioFormats is one such tool, where the XML stored in the ImageDescription tag is explicitly encoded as UTF-8. On Python 3, pytiff.Tiff._read_ascii raises a UnicodeDecodeError upon reading such values. On Python 2, where the treatment of string encoding/decoding isn't as rigorous, the problem is effectively ignored.

Would you accept a patch to fix this? I think the simplest approach is to always decode strings as UTF-8, perhaps only under Python 3. This actually mirrors the way _set_tag already performs UTF-8 string encoding, only under Python 3. I would also be willing to implement a more flexible approach with a user-controlled encoding if you think that's a better option.

This issue is related to the one entitled "Missing dependency: Cython", which was closed. I'm very happy to hear about pytiff! It is exactly what I have been looking for, already for some time. At first I installed pytiff as part of my Python 3.6.5 instance (32-bits). When I tried to run it, I got the error "Cython modules not available". Because I read that this problem had been fixed for Python 3.7, I decided to do a fresh install of Python 3.7.4. After installing, I even upgraded pip. Unfortunately, I have to conclude that the issue has not yet been resolved fully. I'm attaching the error messages. I hope this can be resolved, so that I can use pytiff without hesitations as one of the required packages in a package I'm developing myself. pytiff_issue.txt

I have image stacks which I am trying to load into pytiff. However, it looks like it only loads the first slice:

    with pytiff.Tiff(input_fpath) as handle:
        img_array = handle
        print(img_array.shape)  # (15, 15) instead of (15,15,15)