django-large-image 0.10.3

Dynamic tile server in Django built on top of large-image (and GDAL)

🩻 🗺️ django-large-image

django-large-image is an abstraction of large-image for use with django-rest-framework providing viewset mixins for endpoints to work with large images (Cloud Optimized GeoTiffs or medical image formats) in Django. The dynamic tile server provided here prevents the need for preprocessing large images into tile sets for viewing interactively on slippy-maps. Under the hood, large-image applies operations (rescaling, reprojection, image encoding) to create image tiles on-the-fly.

Lightning Talk for 2022 Cloud-Native Geospatial Outreach Event
View slides here

Table of Contents

• Overview
  • Support
  • Features
• Installation
• Usage
  • Example Code
  • Customization
  • Non-Detail ViewSets
  • Styling
  • Converting Images to Pyramidal Tiffs (COGs)
  • Using with django-raster
• Demo App

---

ℹ️ Overview

This package brings Kitware's large-image to Django by providing a set of abstract, mixin API viewset classes that will handle tile serving, fetching metadata from images, and extracting regions of interest.

django-large-image is an installable Django app with a few classes that can be mixed into a Django project (or application)'s drf-based viewsets to provide tile serving endpoints out of the box. Notably, django-large-image is designed to work specifically with FileField interfaces with development being tailored to Kitware's S3FileField. GeoDjango's GDALRaster can also be used by returning GDALRaster.name in the get_path() override.

This package ships with pre-made HTML templates for rendering geospatial image tiles with CesiumJS and non-geospatial image tiles with GeoJS.

Dynamic tile server in Django built on top of large-image (and GDAL)

🤝 Support

django-large-image and the supporting large-image library are developed and maintained by the Data & Analytics group at Kitware, Inc. We work with large image data in both the geospatial and medical capacities. If you have questions about these technologies, or you would like to discuss your own geospatial and medical image problems and learn how we can help, please reach out at [e-mail]. We look forward to the conversation!

🌟 Features

Rich set of RESTful endpoints to extract information from large image formats:

• Image metadata (/info/metadata, /info/metadata_internal)
• Tile serving (/tiles/{z}/{x}/{y}.png?projection=EPSG:3857)
• Region extraction (/data/region.tif?left=v&right=v&top=v&bottom=v)
• Image thumbnails (/data/thumbnail.png)
• Individual pixels (/data/pixel?left=v&top=v)
• Band histograms (/data/histogram)

Support for any storage backend:

• Supports Django's FileField
• Supports S3FileField
• Customizable method for handling data access (get_path override)
• Supports GDAL's Virtual File System for s3://, ftp://, etc. URLs

Miscellaneous:

• Admin interface widget for viewing image tiles.
• Caching
  • image tiles and thumbnails are cached to prevent recreating these data on multiple requests
  • utilizes the Django cache framework. Specify a named cache to use with the LARGE_IMAGE_CACHE_NAME setting.
• Easily extensible SSR templates for tile viewing with CesiumJS and GeoJS
• OpenAPI specification

OpenAPI Documentation	Tiles Endpoint

⬇️ Installation

Out of the box, django-large-image only depends on the core large-image module, but you will need a large-image-source-* module in order for this to work. Most of our users probably want to work with geospatial images so we will focus on the large-image-source-gdal and ``large-image-source-rasteriocases, but it is worth noting thatlarge-image` has source modules for a wide variety of image formats (e.g., medical image formats for microscopy).

See large-image's installation instructions for more details.

🎡 pip

Rasterio

pip install \
  django-large-image \
  'large-image[rasterio,pil]>=1.22'

GDAL

Tip: installing GDAL is notoriously difficult, so at Kitware we provide pre-built Python wheels with the GDAL binary bundled for easily installation in production linux environments. To install our GDAL wheel, use: pip install --find-links https://girder.github.io/large_image_wheels GDAL

pip install \
  --find-links https://girder.github.io/large_image_wheels \
  django-large-image \
  'large-image[gdal,pil]>=1.16.2'

🐍 Conda

Or install with conda:

conda install -c conda-forge django-large-image large-image-source-rasterio

conda install -c conda-forge django-large-image large-image-source-gdal

🚀 Usage

Simply install the app and mixin one of the mixing classes to your existing django-rest-framework viewset.

# settings.py
INSTALLED_APPS = [
    ...,
    'django_large_image',
]

The following are the provided mixin classes and their use case:

• LargeImageMixin: for use with a standard, non-detail ViewSet. Users must implement get_path()
• LargeImageDetailMixin: for use with a detail viewset like GenericViewSet. Users must implement get_path()
• LargeImageFileDetailMixin: (most commonly used) for use with a detail viewset like GenericViewSet where the associated model has a FileField storing the image data.
• LargeImageVSIFileDetailMixin: (geospatial) for use with a detail viewset like GenericViewSet where the associated model has a FileField storing the image data that is intended to be read with GDAL/rasterio. This will access the data over GDAL's Virtual File System interface (a VSI path).

Most users will want to use LargeImageFileDetailMixin and so the following example demonstrate how to use it:

Specify the FILE_FIELD_NAME as the string name of the FileField in which your image data are saved on the associated model.

# viewsets.py
from rest_framework import viewsets
from django_large_image.rest import LargeImageFileDetailMixin

class MyModelViewSet(viewsets.GenericViewSet, LargeImageFileDetailMixin):
  ...  # configuration for your model's viewset
  FILE_FIELD_NAME = 'field_name'

# urls.py
from django.urls import include, path
from rest_framework.routers import SimpleRouter

from myapp.viewsets import MyModelViewSet

router = SimpleRouter(trailing_slash=False)
router.register(r'api/my-model', MyModelViewSet)

urlpatterns = [
  # Additional, standalone URLs from django-large-image
  path('', include('django_large_image.urls')),
] + router.urls

Optional Cache Configuration

Large Image leverages a cache to optimize performance (see Large Image caching docs). By default, django-large-image will use a Django-based cache. You may also use certain settings to customize the caching behavior and/or use a different cache backend.

The Large Image configuration options specify several cache options, all prefixed with cache_. To set any of these options, specify them in the settings of your Django app as all-caps attributes prefixed with LARGE_IMAGE_. For example, to set Redis as your cache backend, you may specify the following settings:

LARGE_IMAGE_CACHE_BACKEND = "redis"
LARGE_IMAGE_CACHE_REDIS_URL = [redis_url]
LARGE_IMAGE_CACHE_REDIS_USERNAME = [redis_username]
LARGE_IMAGE_CACHE_REDIS_PASSWORD = [redis_password]

In addition to the configuration options listed in the docs linked above, one additional configuration option exists to set the name of the cache. You may set LARGE_IMAGE_CACHE_NAME as any string value. Otherwise, the name will simply be "default".

And that's it!

📝 Example Code

To use the mixin classes provided here, add django_large_image to the INSTALLED_APPS of your Django project, then create a model, serializer, and viewset in your Django project like so:

# models.py
from django.db import models
from rest_framework import serializers


class ImageFile(models.Model):
    name = models.TextField()
    file = models.FileField()


class ImageFileSerializer(serializers.ModelSerializer):
    class Meta:
        model = ImageFile
        fields = '__all__'

# admin.py
from django.contrib import admin
from example.core.models import ImageFile


@admin.register(ImageFile)
class ImageFileAdmin(admin.ModelAdmin):
    list_display = ('pk', 'name')

Then create the viewset, mixing in the django-large-image viewset class:

# viewsets.py
from example.core import models
from rest_framework import mixins, viewsets

from django_large_image.rest import LargeImageFileDetailMixin


class ImageFileDetailViewSet(
    mixins.ListModelMixin,
    viewsets.GenericViewSet,
    LargeImageFileDetailMixin,
):
    queryset = models.ImageFile.objects.all()
    serializer_class = models.ImageFileSerializer

    # for `django-large-image`: the name of the image FileField on your model
    FILE_FIELD_NAME = 'file'

Then register the URLs:

# urls.py
from django.urls import include, path
from example.core.viewsets import ImageFileDetailViewSet
from rest_framework.routers import SimpleRouter

router = SimpleRouter(trailing_slash=False)
router.register(r'api/image-file', ImageFileDetailViewSet)

urlpatterns = [
  # Additional, standalone URLs from django-large-image
  path('', include('django_large_image.urls')),
] + router.urls

(Optional) You can also use an admin widget for your model:

<!-- templates/admin/myapp/imagefile/change_form.html -->
{% extends "admin/change_form.html" %}

{% block after_field_sets %}

<script>
  var baseEndpoint = 'api/image-file';
</script>

{% include 'admin/django_large_image/_include/geojs.html' %}

{% endblock %}

Please note the example Django project in the project/ directory of this repository that shows how to use django-large-image in a girder-4 project.

🛠️ Customization

The mixin classes are modularly designed and able to be subclassed for your project's needs. While the provided LargeImageFileDetailMixin handles FileField-interfaces, you can easily extend its base class, LargeImageDetailMixin, to handle any mechanism of data storage in your detail-oriented viewset.

In the following example, we demonstrate how to use GDAL compatible VSI paths from a model that stores s3:// or https:// URLs.

# models.py
from django.db import models
from rest_framework import serializers


class URLImageFile(models.Model):
    name = models.TextField()
    url = models.TextField()


class URLImageFileSerializer(serializers.ModelSerializer):
    class Meta:
        model = URLImageFile
        fields = '__all__'

# admin.py
from django.contrib import admin
from example.core.models import URLImageFile


@admin.register(URLImageFile)
class URLImageFileAdmin(admin.ModelAdmin):
    list_display = ('pk', 'name')

# viewsets.py
from example.core import models
from rest_framework import mixins, viewsets

from django_large_image.rest import LargeImageDetailMixin
from django_large_image.utilities import make_vsi


class URLLargeImageMixin(LargeImageDetailMixin):
    def get_path(self, request, pk=None):
        object = self.get_object()
        return make_vsi(object.url)


class URLImageFileDetailViewSet(
    mixins.ListModelMixin,
    viewsets.GenericViewSet,
    URLLargeImageMixin,
):
    queryset = models.URLImageFile.objects.all()
    serializer_class = models.URLImageFileSerializer

Here is a good test image: https://oin-hotosm.s3.amazonaws.com/59c66c5223c8440011d7b1e4/0/7ad397c0-bba2-4f98-a08a-931ec3a6e943.tif

🥸 Non-Detail ViewSets

The LargeImageMixin provides a mixin interface for non-detail viewsets (no associated model or primary key required). This can be particularly useful if your viewset has custom logic to retrieve the desired data.

For example, you may want a viewset that gets the data path as a URL embedded in the request's query parameters. To do this, you can make a standard ViewSet with the LargeImageMixin like so:

# viewsets.py
from rest_framework import viewsets
from rest_framework.exceptions import ValidationError

from django_large_image.rest import LargeImageMixin
from django_large_image.utilities import make_vsi


class URLLargeImageViewSet(viewsets.ViewSet, LargeImageMixin):
    def get_path(self, request, pk=None):
        try:
            url = request.query_params.get('url')
        except KeyError:
            raise ValidationError('url must be defined as a query parameter.')
        return make_vsi(url)

🪄 Styling

django-large-image's dynamic tile serving supports band styling and making composite images from multiple frames and/or bands of your images. This means that you can easily create a false color image from multispectral imagery.

django-large-image has two styling modes:

1. A simple interface to colormap a single channel using multiple query parameters. These are the documented OpenAPI query parameters.

View a single band with a Matplotlib colormap:

var thumbnailUrl = `http://localhost:8000/api/image-file/${imageId}/data/thumbnail.png?band=3&palette=viridis&min=50&max=250`;

2. A complex specification for styling across frames and bands to create composite images using a JSON specification defined by large-image.

Create a false color image from multiple bands in the source image:

// See https://girder.github.io/large_image/tilesource_options.html#style
var style = {
  bands: [
    {band: 5, palette: ['#000', '#f00']},  // red
    {band: 3, palette: ['#000', '#0f0']},  // green
    {band: 2, palette: ['#000', '#00f']}   // blue
  ]
};
var styleEncoded = encodeURIComponent(JSON.stringify(style))
var thumbnailUrl = `http://localhost:8000/api/image-file/${imageId}/data/thumbnail.png?style=${styleEncoded}`;

☁️ Converting Images to Pyramidal Tiffs (COGs)

Install large_image_converter and run the following:

import large_image_converter
large_image_converter.convert(input_path, output_path)

It's that easy! The default parameters for that function will convert geospatial rasters to Cloud Optimized GeoTiffs (COGs) and non-geospatial images to a pyramidal tiff format.

It's quite common to have a celery task that converts an image from a model in your application. Here is a starting point:

import os
from example.core import models
from celery import shared_task
import large_image_converter  # requires large-image-source-gdal


@shared_task
def task_convert_cog(my_model_pk):
    image_file = models.ImageFile.objects.get(pk=my_model_pk)
    input_path = image_file.file.name  # TODO: get full path to file on disk

    with tempfile.TemporaryDirectory() as tmpdir:
        output_path = os.path.join(tmpdir, 'converted.tiff')
        large_image_converter.convert(input_path, output_path)

        # Do something with converted tiff file at `output_path`
        ...

If using the rasterio-based source module, we recommend using rio-cogeo over large_image_converter.

Using with django-raster

django-raster is a popular choice for storing geospatial raster data in Django. django-large-image works well with django-raster to provide additional endpoints for dynamic tile serving and more.

Please take a look at the demo project here: https://github.com/ResonantGeoData/django-raster-demo and raise any questions about usage with django-raster there.

Demo App

There is a vanilla Django project in the demo/ directory and this app is published as a standalone Docker image that anyone can try out:

docker run -it -p 8000:8000 -v dli_demo_data:/opt/django-project/data ghcr.io/girder/django-large-image-demo:latest