sij/khoj
1
0
Fork 0
mirror of https://github.com/khoj-ai/khoj.git synced 2024-11-23 15:38:55 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 5

Get XMP metadata from image using Pillow. Remove ExifTool dependency

Browse source 
- Pillow already supports reading XMP metadata from Images
- Removes need to maintain my fork of unmaintained PyExiftool
  - This also removes dependency on system Exiftool package for
    XMP metadata extraction
- Add test to verify XMP metadata extracted from test images
- Remove references to Exiftool from Documentation
This commit is contained in:
Debanjum Singh Solanky 2022-09-14 13:22:27 +03:00
parent 8f57a62675
commit bf1ae038cb
10 changed files with 35 additions and 350 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
Dockerfile
View file

@ -4,7 +4,7 @@ LABEL org.opencontainers.image.source https://github.com/debanjum/khoj
# Install System Dependencies
RUN apt-get update -y && \
apt-get -y install libimage-exiftool-perl python3-pyqt5
apt-get -y install python3-pyqt5
# Copy Application to Container
COPY . /app

7
Readme.md
View file

@ -208,11 +208,7 @@ docker-compose build --pull
#### Using Conda
##### 1. Install Dependencies
- [Install Conda](https://docs.conda.io/projects/conda/en/latest/user-guide/install/index.html) \[Required\]
- Install Exiftool \[Optional\]
``` shell
sudo apt -y install libimage-exiftool-perl
```
- [Install Conda](https://docs.conda.io/projects/conda/en/latest/user-guide/install/index.html)
##### 2. Install Khoj
```shell
@ -254,4 +250,3 @@ pytest
- Charles Cave for [OrgNode Parser](http://members.optusnet.com.au/~charles57/GTD/orgnode.html)
- [Org.js](https://mooz.github.io/org-js/) to render Org-mode results on the Web interface
- [Markdown-it](https://github.com/markdown-it/markdown-it) to render Markdown results on the Web interface
- Sven Marnach for [PyExifTool](https://github.com/smarnach/pyexiftool/blob/master/exiftool.py)

1
setup.py
View file

@ -40,6 +40,7 @@ setup(
"aiofiles == 0.8.0",
"dateparser == 1.1.1",
"pyqt6 == 6.3.1",
"defusedxml == 0.7.1",
],
include_package_data=True,
entry_points={"console_scripts": ["khoj = src.main:run"]},

12
src/search_type/image_search.py
View file

@ -13,8 +13,7 @@ from tqdm import trange
import torch
# Internal Packages
from src.utils.helpers import get_absolute_path, resolve_absolute_path, load_model
import src.utils.exiftool as exiftool
from src.utils.helpers import get_absolute_path, get_from_dict, resolve_absolute_path, load_model
from src.utils.config import ImageSearchModel
from src.utils.rawconfig import ImageContentConfig, ImageSearchConfig
@ -122,11 +121,12 @@ def compute_metadata_embeddings(image_names, encoder, embeddings_file, batch_siz
def extract_metadata(image_name):
with exiftool.ExifTool() as et:
image_metadata = et.get_tags(["XMP:Subject", "XMP:Description"], str(image_name))
image_metadata_subjects = set([subject.split(":")[1] for subject in image_metadata.get("XMP:Subject", "") if ":" in subject])
image_xmp_metadata = Image.open(image_name).getxmp()
image_description = get_from_dict(image_xmp_metadata, 'xmpmeta', 'RDF', 'Description', 'description', 'Alt', 'li', 'text')
image_subjects = get_from_dict(image_xmp_metadata, 'xmpmeta', 'RDF', 'Description', 'subject', 'Bag', 'li')
image_metadata_subjects = set([subject.split(":")[1] for subject in image_subjects if ":" in subject])
image_processed_metadata = image_metadata.get("XMP:Description", "")
image_processed_metadata = image_description
if len(image_metadata_subjects) > 0:
image_processed_metadata += ". " + ", ".join(image_metadata_subjects)

325
src/utils/exiftool.py
View file

@ -1,325 +0,0 @@
# -*- coding: utf-8 -*-
# PyExifTool <http://github.com/smarnach/pyexiftool>
# Copyright 2012 Sven Marnach
# This file is part of PyExifTool.
#
# PyExifTool is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the licence, or
# (at your option) any later version, or the BSD licence.
#
# PyExifTool is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
#
# See COPYING.GPL or COPYING.BSD for more details.
"""
PyExifTool is a Python library to communicate with an instance of Phil
Harvey's excellent ExifTool_ command-line application. The library
provides the class :py:class:`ExifTool` that runs the command-line
tool in batch mode and features methods to send commands to that
program, including methods to extract meta-information from one or
more image files. Since ``exiftool`` is run in batch mode, only a
single instance needs to be launched and can be reused for many
queries. This is much more efficient than launching a separate
process for every single query.
.. _ExifTool: http://www.sno.phy.queensu.ca/~phil/exiftool/
The source code can be checked out from the github repository with
::
git clone git://github.com/smarnach/pyexiftool.git
Alternatively, you can download a tarball_. There haven't been any
releases yet.
.. _tarball: https://github.com/smarnach/pyexiftool/tarball/master
PyExifTool is licenced under GNU GPL version 3 or later.
Example usage::
import exiftool
files = ["a.jpg", "b.png", "c.tif"]
with exiftool.ExifTool() as et:
metadata = et.get_metadata_batch(files)
for d in metadata:
print("{:20.20} {:20.20}".format(d["SourceFile"],
d["EXIF:DateTimeOriginal"]))
"""
from __future__ import unicode_literals
import sys
import subprocess
import os
import json
import warnings
import codecs
try: # Py3k compatibility
basestring
except NameError:
basestring = (bytes, str)
executable = "exiftool"
"""The name of the executable to run.
If the executable is not located in one of the paths listed in the
``PATH`` environment variable, the full path should be given here.
"""
# Sentinel indicating the end of the output of a sequence of commands.
# The standard value should be fine.
sentinel = b"{ready}"
# The block size when reading from exiftool. The standard value
# should be fine, though other values might give better performance in
# some cases.
block_size = 4096
# This code has been adapted from Lib/os.py in the Python source tree
# (sha1 265e36e277f3)
def _fscodec():
encoding = sys.getfilesystemencoding()
errors = "strict"
if encoding != "mbcs":
try:
codecs.lookup_error("surrogateescape")
except LookupError:
pass
else:
errors = "surrogateescape"
def fsencode(filename):
"""
Encode filename to the filesystem encoding with 'surrogateescape' error
handler, return bytes unchanged. On Windows, use 'strict' error handler if
the file system encoding is 'mbcs' (which is the default encoding).
"""
if isinstance(filename, bytes):
return filename
else:
return filename.encode(encoding, errors)
return fsencode
fsencode = _fscodec()
del _fscodec
class ExifTool(object):
"""Run the `exiftool` command-line tool and communicate to it.
You can pass the file name of the ``exiftool`` executable as an
argument to the constructor. The default value ``exiftool`` will
only work if the executable is in your ``PATH``.
Most methods of this class are only available after calling
:py:meth:`start()`, which will actually launch the subprocess. To
avoid leaving the subprocess running, make sure to call
:py:meth:`terminate()` method when finished using the instance.
This method will also be implicitly called when the instance is
garbage collected, but there are circumstance when this won't ever
happen, so you should not rely on the implicit process
termination. Subprocesses won't be automatically terminated if
the parent process exits, so a leaked subprocess will stay around
until manually killed.
A convenient way to make sure that the subprocess is terminated is
to use the :py:class:`ExifTool` instance as a context manager::
with ExifTool() as et:
...
.. warning:: Note that there is no error handling. Nonsensical
options will be silently ignored by exiftool, so there's not
much that can be done in that regard. You should avoid passing
non-existent files to any of the methods, since this will lead
to undefied behaviour.
.. py:attribute:: running
A Boolean value indicating whether this instance is currently
associated with a running subprocess.
"""
def __init__(self, executable_=None):
if executable_ is None:
self.executable = executable
else:
self.executable = executable_
self.running = False
def start(self):
"""Start an ``exiftool`` process in batch mode for this instance.
This method will issue a ``UserWarning`` if the subprocess is
already running. The process is started with the ``-G`` and
``-n`` as common arguments, which are automatically included
in every command you run with :py:meth:`execute()`.
"""
if self.running:
warnings.warn("ExifTool already running; doing nothing.")
return
with open(os.devnull, "w") as devnull:
self._process = subprocess.Popen(
[self.executable, "-stay_open", "True", "-@", "-",
"-common_args", "-G", "-n"],
stdin=subprocess.PIPE, stdout=subprocess.PIPE,
stderr=devnull)
self.running = True
def terminate(self):
"""Terminate the ``exiftool`` process of this instance.
If the subprocess isn't running, this method will do nothing.
"""
if not self.running:
return
self._process.stdin.write(b"-stay_open\nFalse\n")
self._process.stdin.flush()
self._process.communicate()
del self._process
self.running = False
def __enter__(self):
self.start()
return self
def __exit__(self, exc_type, exc_val, exc_tb):
self.terminate()
def __del__(self):
self.terminate()
def execute(self, *params):
"""Execute the given batch of parameters with ``exiftool``.
This method accepts any number of parameters and sends them to
the attached ``exiftool`` process. The process must be
running, otherwise ``ValueError`` is raised. The final
``-execute`` necessary to actually run the batch is appended
automatically; see the documentation of :py:meth:`start()` for
the common options. The ``exiftool`` output is read up to the
end-of-output sentinel and returned as a raw ``bytes`` object,
excluding the sentinel.
The parameters must also be raw ``bytes``, in whatever
encoding exiftool accepts. For filenames, this should be the
system's filesystem encoding.
.. note:: This is considered a low-level method, and should
rarely be needed by application developers.
"""
if not self.running:
raise ValueError("ExifTool instance not running.")
self._process.stdin.write(b"\n".join(params + (b"-execute\n",)))
self._process.stdin.flush()
output = b""
fd = self._process.stdout.fileno()
while not output[-32:].strip().endswith(sentinel):
output += os.read(fd, block_size)
return output.strip()[:-len(sentinel)]
def execute_json(self, *params):
"""Execute the given batch of parameters and parse the JSON output.
This method is similar to :py:meth:`execute()`. It
automatically adds the parameter ``-j`` to request JSON output
from ``exiftool`` and parses the output. The return value is
a list of dictionaries, mapping tag names to the corresponding
values. All keys are Unicode strings with the tag names
including the ExifTool group name in the format <group>:<tag>.
The values can have multiple types. All strings occurring as
values will be Unicode strings. Each dictionary contains the
name of the file it corresponds to in the key ``"SourceFile"``.
The parameters to this function must be either raw strings
(type ``str`` in Python 2.x, type ``bytes`` in Python 3.x) or
Unicode strings (type ``unicode`` in Python 2.x, type ``str``
in Python 3.x). Unicode strings will be encoded using
system's filesystem encoding. This behaviour means you can
pass in filenames according to the convention of the
respective Python version – as raw strings in Python 2.x and
as Unicode strings in Python 3.x.
"""
params = map(fsencode, params)
return json.loads(self.execute(b"-j", *params).decode("utf-8"))
def get_metadata_batch(self, filenames):
"""Return all meta-data for the given files.
The return value will have the format described in the
documentation of :py:meth:`execute_json()`.
"""
return self.execute_json(*filenames)
def get_metadata(self, filename):
"""Return meta-data for a single file.
The returned dictionary has the format described in the
documentation of :py:meth:`execute_json()`.
"""
return self.execute_json(filename)[0]
def get_tags_batch(self, tags, filenames):
"""Return only specified tags for the given files.
The first argument is an iterable of tags. The tag names may
include group names, as usual in the format <group>:<tag>.
The second argument is an iterable of file names.
The format of the return value is the same as for
:py:meth:`execute_json()`.
"""
# Explicitly ruling out strings here because passing in a
# string would lead to strange and hard-to-find errors
if isinstance(tags, basestring):
raise TypeError("The argument 'tags' must be "
"an iterable of strings")
if isinstance(filenames, basestring):
raise TypeError("The argument 'filenames' must be "
"an iterable of strings")
params = ["-" + t for t in tags]
params.extend(filenames)
return self.execute_json(*params)
def get_tags(self, tags, filename):
"""Return only specified tags for a single file.
The returned dictionary has the format described in the
documentation of :py:meth:`execute_json()`.
"""
return self.get_tags_batch(tags, [filename])[0]
def get_tag_batch(self, tag, filenames):
"""Extract a single tag from the given files.
The first argument is a single tag name, as usual in the
format <group>:<tag>.
The second argument is an iterable of file names.
The return value is a list of tag values or ``None`` for
non-existent tags, in the same order as ``filenames``.
"""
data = self.get_tags_batch([tag], filenames)
result = []
for d in data:
d.pop("SourceFile")
result.append(next(iter(d.values()), None))
return result
def get_tag(self, tag, filename):
"""Extract a single tag from a single file.
The return value is the value of the specified tag, or
``None`` if this tag was not found in the file.
"""
return self.get_tag_batch(tag, [filename])[0]

BIN
tests/data/images/guineapig_grass.jpg
View file

Binary file not shown.
Side by side Swipe Overlay

Before

Width:  |  Height:  |  Size: 170 KiB

After

Width:  |  Height:  |  Size: 173 KiB

BIN
tests/data/images/horse_dog.jpg
View file

Binary file not shown.
Side by side Swipe Overlay

Before

Width:  |  Height:  |  Size: 330 KiB

After

Width:  |  Height:  |  Size: 334 KiB

BIN
tests/data/images/kitten_park.jpg
View file

Binary file not shown.
Side by side Swipe Overlay

Before

Width:  |  Height:  |  Size: 268 KiB

After

Width:  |  Height:  |  Size: 271 KiB

8
tests/data/markdown/main_readme.md
View file

@ -96,12 +96,6 @@ docker-compose build --pull
Conda](https://docs.conda.io/projects/conda/en/latest/user-guide/install/index.html)
\[Required\]
3. Install Exiftool \[Optional\]
``` shell
sudo apt-get -y install libimage-exiftool-perl
```
2. 2\. Install Khoj
``` shell
@ -149,5 +143,3 @@ conda activate khoj
Documentation](https://www.sbert.net/examples/applications/image-search/README.html)
- Charles Cave for [OrgNode
Parser](http://members.optusnet.com.au/~charles57/GTD/orgnode.html)
- Sven Marnach for
[PyExifTool](https://github.com/smarnach/pyexiftool/blob/master/exiftool.py)

22
tests/test_image_search.py
View file

@ -25,6 +25,28 @@ def test_image_search_setup(content_config: ContentConfig, search_config: Search
assert len(image_search_model.image_embeddings) == 3
def test_image_metadata(content_config: ContentConfig):
"Verify XMP Description and Subjects Extracted from Image"
# Arrange
expected_metadata_image_name_pairs = [
(["Billi Ka Bacha.", "Cat", "Grass"], "kitten_park.jpg"),
(["Pasture.", "Horse", "Dog"], "horse_dog.jpg"),
(["Guinea Pig Eating Celery.", "Rodent", "Whiskers"], "guineapig_grass.jpg")]
test_image_paths = [
Path(content_config.image.input_directories[0] / image_name[1])
for image_name in expected_metadata_image_name_pairs
]
for expected_metadata, test_image_path in zip(expected_metadata_image_name_pairs, test_image_paths):
# Act
actual_metadata = image_search.extract_metadata(test_image_path)
# Assert
for expected_snippet in expected_metadata[0]:
assert expected_snippet in actual_metadata
# ----------------------------------------------------------------------------------------------------
def test_image_search(content_config: ContentConfig, search_config: SearchConfig):
# Arrange