Original issue noted here: https://github.com/ruven/iipsrv/issues/98 but copied below since this is really a vips issue, not an IIP issue.

I have confirmed that there is an ongoing resize bug in vips that is reproducible on the latest up-to-date RHEL6 and RHEL7 systems using either vips 7.x or the latest 8.4 version of vips. See the differences between the images below which were taken from the pyramidal tiff. The first was created using imagemagick and the second with vips 8.4.5 on RHEL7. Aside from being slower, we had found a few years back that IM creates intermittent corrupted ptifs that IIP can't open. I think the only option is to find the bug in vips or the library it's using and fix it - or switch to IM if the PTIF generation has been fixed since we last looked at it about 4 or 5 years ago. Any advice appreciated.

One of the resolutions from an ImageMagick generated pyramidal tiff image

and same resolution from vips generated pyramidal tiff (note flatness on right side and bottom) image

command used to generate the ptif: vips im_vips2tiff ./source_tiff_file.tif vips_regen_8.4.5_rhel7.ptif:jpeg:90,tile:256x256,pyramid

info about the original TIFF source: TIFF 11038x11038 11038x11038+0+0 8-bit DirectClass 365.5MB

command used to generate the output above

convert vips_regen_8.4.5_rhel7.ptif -quality 100 vips_regen_8.4.5_rhel7.jpg

This produces one jpg for each of the 6 or 7 resolutions that vips generates in the ptif. All of the exported jpegs from a ptif generated by vips exhibit this problem whereas none of the exported jpgs generated with imagemagick do.

I'm new to libvips and want to perform common image corrections like the adjustment of white balance, hue, brightness or tonal range. Are there tutorials/documentations covering these adjustments? I'm also interested in applying filters like gaussian blur etc. with libvips

Hi i have a folder with these images: image_1.png, image_2.png, image_3.png. Its possible to join all this images togheter horizontaly? I found vips join, but that can only merge 2 images.

In ImageMagick i would do: convert +append image_*.png output.png

Thanks.

Let's get 8.9.0 out of the door now that streams are merged.

This issue is for tracking the final checks before release.

• [x] test win builds
• [x] test macos
• [x] write a "what's new" post
• [x] test ruby-vips
• [x] test pyvips
• [x] test php-vips
• [x] test lua-vips
• [x] clear out PRs
• [x] clear out bug issues
• [x] add a thing to make old-style file and buffer loaders usable from the stream API
• [x] benchmark
• [x] out of order read in tiffload?
• [x] decide: should we revert sharpen behaviour?
• [x] update API docs

@lovell and @kleisauke, do you have any things you'd like to add for 8.9?

Hello,

I have seen this conda package for libvips: https://anaconda.org/zegami/libvips

However, it has some limitations (e.g. openslide is not included).

Is there anyone trying to build a conda package for libvips, and make it available either through conda-forge or bioconda?

Best regards, Sebastian

Background: I have a Rust web server that do some image processing using OpenCV and I want to switch to libvips mostly because of performance.

Lately I've been trying to write bindings for the libvips C library for rust and I'd like to start with the things I have to support in my application, which are decoding images from any known format, resizing, rotating, watermarking and encoding to jpeg or webp.

I've been reading the docs and I've got a good idea on how the library works, but still got a couple doubts. Please bear in mind that I'm not a very experienced C programmer. I have some experience with C++, but still it's been almost 10 years since last time I used it.

Here it goes:

1. For the functions that take varargs options (the option named parameters), how do I specify them? Are them strings and should be specified like the options in the read from file options where you can just pass in the file name the options like this file.jpeg[shrink=2]? So should I specify a list with of option=value as strings (const *char)? Still about this options, how should I specify the ones that are enums?

2. What are the types for the PNG options in the save function? https://libvips.github.io/libvips/API/current/VipsForeignSave.html#vips-pngsave Is there a place where I can find the defaults for the various options when encoding and decoding images?

3. One of the features I have to support envolves removing the alpha channel of images when encoding to JPEG. I know it works out of the box, but I want a different behavior. The default way, it transform the transparent pixels to black, which makes sense, since 0 = transparent = black. How can I invert the alpha channel in a image with libvips to get the opposite result (the fully transparent pixels getting white instead of black)?

4. About the error handling... How should I proceed? Should I have a background thread consuming and clearing the error buffer from time to time? If so how would it be a reasonable time to do it without degrading the performance of my application (because of the locks)...

5. Still about error handling, if I call vips_error_freeze, the buffer will never be filled, meaning I never have to care clearing it, right?

6. For watermarking I found out I should use the vips_composite2 function with the VIPS_BLEND_MODE_OVERLAY mode, but how can I set an opacity for the overlay image?

Thank you in advance.

I run a program that stitches images together to an overview using merge() and join(), performs some operations on the overview image, and writes the resulting image to disk in form of a zoomify pyramid using dzsave().

Individual images exists on disk as PNG files with a size of 3.5 to 6MB per file. The program works fine with small overviews, i.e. small number of images, but fails with a large number of images (about 10000 files).

The program get's killed with exitcode 137 (128+9) after the stitching as soon as a function is applied to the resulting Vips.Image object.

The error message is not very informative, it just says Killed.

This is the tail of the DEBUG log output:

2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | _call_base self=<Image object at 0x7f1f0ea0dc80 (VipsImage at 0xe9e687e0)>
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | assigning <Image object at 0x7f1f0ea0dc80 (VipsImage at 0xe9e687e0)> to in1
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | in1 needs a <GType VipsImage (33612672)>
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | assigning <Image object at 0x7f1f0ea0dc80 (VipsImage at 0xe9e687e0)>
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | assigning <Image object at 0x7f1f0fcc2eb0 (VipsImage at 0x2ee0ca0)> to in2
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | in2 needs a <GType VipsImage (33612672)>
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | assigning <Image object at 0x7f1f0fcc2eb0 (VipsImage at 0x2ee0ca0)>
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | assigning vertical to direction
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | direction needs a <GType VipsDirection (37071744)>
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | assigning vertical
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | _call_base checking cache for op <__main__.VipsJoin object at 0x7f1f0ccaf3c0 (VipsJoin at 0xe9cc4e50)>
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | _call_base got op2 <__main__.VipsJoin object at 0x7f1f0ccaf3c0 (VipsJoin at 0xe9cc4e50)>
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | _call_base fetching required output args
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | read out <Image object at 0x7f1f0f1b88c0 (VipsImage at 0xe9fdaaf0)> from out
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | _call_base fetching optional output args
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | success
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | Image.__getattr__ percent
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | _call_base name=percent, required=(99.9,) optional={}
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | _call_base self=<Image object at 0x7f1f0f1b88c0 (VipsImage at 0xe9fdaaf0)>
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | assigning <Image object at 0x7f1f0f1b88c0 (VipsImage at 0xe9fdaaf0)> to in
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | in needs a <GType VipsImage (33612672)>
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | assigning <Image object at 0x7f1f0f1b88c0 (VipsImage at 0xe9fdaaf0)>
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | assigning 99.9 to percent
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | percent needs a <GType gdouble (60)>
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | assigning 99.9
2015-12-08 16:05:27 | gi.overrides.Vips              | DEBUG    | _call_base checking cache for op <__main__.VipsPercent object at 0x7f1f0f1b8910 (VipsPercent at 0xe9e6e980)>
Killed

The program was run on a machine with 32GB memory and 8 cores.

The same error occurs when another function is called, e.g. scale() instead of percent().

Here is a small test program that illustrates the actual program and might be helpful to reproduce the error:

#! /usr/bin/env python
import os
import numpy as np
import shutil
import tempfile
import itertools
import logging
import argparse
from gi.repository import Vips


def configure_logging():
    logger = logging.getLogger()

    fmt = '%(asctime)s | %(name)-30s | %(levelname)-8s | %(message)s'
    datefmt = '%Y-%m-%d %H:%M:%S'
    formatter = logging.Formatter(fmt=fmt, datefmt=datefmt)

    handler = logging.StreamHandler()
    handler.name = 'stream'
    handler.setFormatter(formatter)
    logger.addHandler(handler)


if __name__ == '__main__':

    configure_logging()

    parser = argparse.ArgumentParser('Test program for Vips pyramid creation.')
    parser.add_argument('-v', '--verbosity', action='count', default=0,
                        help='increase logging verbosity')
    parser.add_argument('-m', '--mosaic_axis_length', default=2, type=int,
                        help='number of images along each dimension of a mosaic')
    parser.add_argument('-o', '--overview_axis_length', default=5, type=int,
                        help='number of mosaics along each dimension of the overview')

    tmp_dir = tempfile.gettempdir()

    args = parser.parse_args()

    logging_level_mapper = {
        0: logging.CRITICAL,
        1: logging.INFO,
        2: logging.DEBUG,
    }

    logger = logging.getLogger('test_vips_pyramid')
    logger.setLevel(logging_level_mapper[args.verbosity])

    vips_logger = logging.getLogger('gi.overrides.Vips')
    vips_logger.setLevel(logging_level_mapper[args.verbosity])

    # Create a squared "overview" image that is composed of many smaller squared
    # "mosaics", each of which is composed of several individual images.
    # NOTE: The size of the actual overview image is at least 0.5 * 10^6 pixels
    # along each dimension.
    mosaic_axis_length = args.mosaic_axis_length
    files = ['file_%d.png' % i for i in range(mosaic_axis_length**2)]

    overview_axis_length = args.overview_axis_length
    folder_names = ['directory_%d' % i for i in range(overview_axis_length**2)]

    input_dir = os.path.join(tmp_dir, 'vips_test', 'inputs')
    if not os.path.exists(input_dir):
        os.makedirs(input_dir)

    output_dir = os.path.join(tmp_dir, 'vips_test', 'outputs')
    if os.path.exists(output_dir):
        shutil.rmtree(output_dir)

    directories = [os.path.join(input_dir, f) for f in folder_names]

    # Create images on disk
    # NOTE: The actual images would already exist on disk
    logger.info('create images')
    for d in directories:
        if not os.path.exists(d):
            os.mkdir(d)
        for f in files:
            filename = os.path.join(input_dir, d, f)
            logger.debug('write image to file: %s', filename)
            img = Vips.Image.gaussnoise(100, 100)
           # NOTE: The actual dimensions are 2500 x 2600
            img.write_to_file(filename)

    logger.info('create mosaics')
    moasic_coordinates = list(itertools.product(
                             range(mosaic_axis_length),
                             range(mosaic_axis_length)))
    mosaics = list()
    for i, d in enumerate(directories):
        logger.debug('stitch mosaic # %d', i)
        # Build grid for the current mosaic
        mosaic_grid_dims = (mosaic_axis_length, mosaic_axis_length)
        mosaic_grid = np.empty(mosaic_grid_dims, dtype='O')
        for i, coords in enumerate(moasic_coordinates):
            filename = os.path.join(d, files[i])
            mosaic_grid[coords[0], coords[1]] = filename

        # Load individual images and stitch them together according to the grid layout
        for r in xrange(mosaic_grid.shape[0]):
            for c in xrange(mosaic_grid.shape[1]):

                img = Vips.Image.new_from_file(mosaic_grid[r, c])

                if c == 0:
                    row = img
                else:
                    row = row.merge(img, 'horizontal', -img.width, 0)

            if r == 0:
                mosaic = row
            else:
                mosaic = mosaic.merge(row, 'vertical', 0, -img.height)

        mosaics.append(mosaic)


    logger.info('create overview')
    overview_coordinates = list(itertools.product(
                              range(overview_axis_length),
                              range(overview_axis_length)))
    # Build grid for the final overview image
    overview_grid_dims = (overview_axis_length, overview_axis_length)
    overiew_grid = np.empty(overview_grid_dims, dtype='O')
    for i, coords in enumerate(overview_coordinates):
        overiew_grid[coords[0], coords[1]] = mosaics[i]

    # Create spacer images, which will be inserted between individual mosaics
    img_dtype = img.get_format()
    col_spacer = Vips.Image.black(10, mosaic.height, bands=1).cast(img_dtype)
    row_width = (
        mosaic.width * overview_grid_dims[1] +
        10 * (overview_grid_dims[1] + 1)
    )
    row_spacer = Vips.Image.black(row_width, 10, bands=1).cast(img_dtype)

    # Stitch mosaics together according to the grid layout
    overview = row_spacer
    for r in xrange(overiew_grid.shape[0]):
        for c in xrange(overiew_grid.shape[1]):

            if c == 0:
                row = col_spacer

            mosaic = overiew_grid[r, c]
            row = row.join(mosaic, 'horizontal')
            row = row.join(col_spacer, 'horizontal')

        overview = overview.join(row, 'vertical')
        overview = overview.join(row_spacer, 'vertical')

    # Clip intensity values
    logger.info('clip intensities')
    threshold = overview.percent(99.9)
    identity_image = Vips.Image.identity(ushort=True)
    condition_image = (identity_image >= threshold)
    lut = condition_image.ifthenelse(threshold, identity_image)
    overview = overview.maplut(lut)

    # Rescale to 8-bit
    logger.info('rescale intensities')
    overview = overview.scale()

    # Save as pyramid
    logger.info('create pyramid: %s', output_dir)
    overview.dzsave(output_dir, layout='zoomify')

Run the program in DEBUG mode:

test_vips_pyramid.py -vv

Vips was installed from git (branch 8.1).

I would like to include @riccardomurri and @smaffiol.

Hi, we get the error vips2png: unable to write to target when converting certain tiff files to png. This error is sporadic. Does this mean that it is unable to write to the file?

Vips: 8.9.0 - Running inside alpine docker image Using bimg to talk to vips

Hi John,

I have run into what seems to be a bug in vips8 or the Ruby interface to it.

I am processing large numbers of 2K tif video images: 2048 x 1556 pixels. For each image: Image file gets read. Modified in some way. Modified image file gets written to a new file.

This was working OK with ruby-vips (vips7). I changed over to vips8 and the gir_ffi-vips interface. All was working OK when processing just a handful of files (testing). When I tried to ramp it up to hundreds of files, there was a crash after about 30 files.

I have isolated this to a very simple example ruby script which just reads a series of numbered 2048 x 1556 pixel all-black tif images and writes each image out — all to the same output file — the error appears when reading a series of different files — writing to different output files or to the same output file makes no difference. This bombs on my system after 16 images. The error message trace back to Vips::Image.write_to_file is:

GLib-GObject-CRITICAL **: g_value_type_compatible: assertion 'G_TYPE_IS_VALUE (src_type)' failed value.rb:158: in check_type_compatibility': Vips::Image is incompatible with VipsImage (ArgumentError) from /Library/Ruby/Gems/2.0.0/gems/gir_ffi-0.7.9/lib/ffi-gobject/value.rb:139:inset_instance_enhanced' from /Library/Ruby/Gems/2.0.0/gems/gir_ffi-0.7.9/lib/ffi-gobject/value.rb:53:in set_value' from /Library/Ruby/Gems/2.0.0/gems/gir_ffi-0.7.9/lib/ffi-gobject/object.rb:77:inset_property_with_override' from /Library/Ruby/Gems/2.0.0/gems/gir_ffi-vips-0.1.0/lib/vips8/argument.rb:130:in set_value' from /Library/Ruby/Gems/2.0.0/gems/gir_ffi-vips-0.1.0/lib/vips8/call.rb:73:incall_base' from /Library/Ruby/Gems/2.0.0/gems/gir_ffi-vips-0.1.0/lib/vips8/image.rb:384:in `write_to_file'

Mostly this is the error I get. One time the error was an address error (a zero address).

Here is the simple test script:

require 'vips8'
require 'pathname'

image_folder = Pathname(__FILE__).dirname + "IMAGE_FOLDER"

(1..50).each do |image_number|

  puts "Image #{image_number}"

  source_file_path = image_folder + "test-2K-image-#{sprintf('%02d', image_number)}.tif"

  test_image = Vips::Image.new_from_file (source_file_path).to_s

  test_image.write_to_file((image_folder + "output_image.tif").to_s)
end

The test files are named "test-2K-image-01", etc. I can't attach a sample file because GitHub doesn't support uploading tif files. Just create a test file in Photoshop then duplicate and number the duplicates.

As I say, the ruby-vips equivalent code worked without a problem.

Any ideas?

Paul Howson

It would be great if libvips had a direct binding to a pdf library to allow efficient generation of eg. thumbnails from PDFs.

vips currently can read pdfs via imagemagick or graphicksmagic, but both of these are very slow and memory inefficient at that.

I think a good candidate could be libpoppler which has a cairo output device (Usage in pdftocairo utility).

We're converting aperio .svs images directly to a google maps pyramid.

In some basic tests, we're seeing a huge variance between running with and without centre.

One of our smaller examples - a ~84 MB .svs:

without

ubuntu@Tortuga:/opt/Processing$ time vips dzsave 2 a --layout google --background black

real     0m43.411s
user 2m44.544s
sys  0m42.400s

with

ubuntu@Tortuga:/opt/Processing$ time vips dzsave 2 b --layout google --background black --centre

real     17m6.623s
user 73m9.596s
sys  17m53.780s

Is this fairly par for the course? Is there anyway we can pre-process the image to optimize conversion?

We have live samples that are 2 GB+ that are taking an hour without the flag, I'm not sure I want to see the difference at that rate. :)

In line with the previous behaviour of GLib's threadpool, which would help in long-running processes. This PR is best reviewed per commit.

Unfortunately, the added internal function (vips__semaphore_down_timed()) was based on vips_semaphore_downn(), so there are currently two copies of almost the same function. I don't think we can safely change vips_semaphore_downn() as that could be called with n > 1, which could conflict with the error code (-1) upon return.

Also, deprecate vips_g_thread_join() in favor of g_thread_join() while we are here (commit 46aee63a4f43e6486867208f5e049f42541eccd7).

Context: https://github.com/libvips/libvips/issues/3240#issuecomment-1367931803.

Note: this PR targets the 8.14 branch.

This branch was successfully deployed on the service associated with the weserv/images project.

We have 8.14.0-rc1 now, including win32 binaries:

https://github.com/libvips/libvips/releases/tag/v8.14.0-rc1

Website updated in this branch:

https://github.com/libvips/libvips/tree/gh-pages-8.14

Draft release notes here:

https://github.com/libvips/libvips/blob/gh-pages-8.14/_posts/2022-12-22-What's-new-in-8.14.md

TODO

• [ ] there have been some minor doc fixes since rc1, we need to regen the C++ binding
• [ ] and the website docs
• [x] test downstream bindings ... pyvips and ruby-vips seem fine
• [x] test sharp
• [ ] test packaging for deb and rpm now that auto* has gone

At the moment, g_warning() messages are sometimes internationalised, and sometimes not. We should pick a policy and apply it to all of libvips.

Suggestion: don't bother.

Some AVIF images fail to be loaded with heifload. For instance: test.avif.zip

Maybe related to https://github.com/libvips/libvips/issues/1808.

$ vips copy test.avif out.jpg

(vips:50621): VIPS-WARNING **: 15:14:52.085: heifload: ignoring nclx profile

(vips:50621): VIPS-WARNING **: 15:14:52.087: heifload: ignoring nclx profile

(vips:50621): VIPS-WARNING **: 15:14:52.088: error in tile 0 x 0

(vips:50621): VIPS-WARNING **: 15:14:52.088: error in tile 0 x 1
[..]
test.avif: bad seek to 2133
heifload: bad image dimensions on decode

vipsheader says the dimensions are 296x196:

$ vipsheader test.avif
test.avif: 296x196 uchar, 3 bands, srgb, heifload

However they should be 296x197 (dimensions of the original JPEG). The AVIF has been created with heif-enc (latest master + SVT-AV1 as encoder):

$ heif-enc -A -e svt test.jpg

test.jpg

libvips version: latest master

Hello :)

Is there a way to read/write the TIFFTAG_SMINSAMPLEVALUE and TIFFTAG_SMAXSAMPLEVALUE tiff tag using libvips? (https://www.awaresystems.be/imaging/tiff/tifftags/sminsamplevalue.html)

I can see that the wtiff_write_header in vips2tiff.c handles a few tiff tags but not all of them.

Thanks for your help,

Lionel