projectile

A tile-on-demand tile server built with PIL and Tornado.

Motivation

We want to store a high-resolution image in memory on the server as a numpy array. Then when a client requests a particular tile, we can make the PNG of the requested tile by slicing the numpy array and using PIL to write the resulting PNG back to the client through a StringIO stream.

This is primarily intended for building interactive visualizations in research settings where we might want to skip the time- and/or disk-intensive tile generation step required by typical tile servers.

Demo

Install projectile

$ pip install projectile

Serve a test image from the USC-SIPI Image Database:

$ projectile sanfran

Manually request a particular tile by navigating to http://localhost:8000/2/1/2.png.

Try zooming and panning in the demo client by navigating to http://localhost:8000/.

Serve one of your own images (any format readable by PIL) with

$ projectile some_image.tiff

or, if you have data in a numpy .npy file,

$ projectile some_image.npy

Load an image in grayscale mode and apply a colormap:

$ projectile --mode L --cmap viridis pentagon

Stress testing

To test the performance limits of on-demand tiling, download this high resolution map of Great Britain (8,150 × 13,086 pixels, file size: 102.74 MB) from Wikimedia Commons.

Grayscale performance test:

$ projectile britain.jpg -m L

Reducing tile resolution when running in color:

$ projectile britain.jpg --tile_size 128

Dependencies

• numpy>=1.13.3
• Pillow>=4.3.0
• tornado>=4.5.2
• matplotlib>=2.1.0

API

URL scheme

The server will serve grayscale and RGB images in their original colors at

/<z>/<x>/<y>/<s>.png

where <z> is the zoom level, <x> and <y> specify the coordinates of the tile at that zoom level (0/0 is the top left tile), and <s> specifies the image tile resolution in pixels (must be a power of 2).

The server will serve colormapped versions of a grayscale image at

/<z>/<x>/<y>/<s>/<cmap>/<vmin>/<vmax>.png

where <cmap> is the name of a matplotlib colormap, and <vmin> and <vmax> specify the range of image pixel values linearly interpolate against the colormap (pixel values outside this range will be clipped).

Using a custom client

If you like the projectile backend but just want to use a simple custom client contained in a single HTML file custom_client.html, you can run

$ projectile array.npy --client custom_client.html

to make projectile serve your client instead of the included demo client.

Using projectile in your existing Tornado web application

The core functionality is exposed in the TileHandler class defined in server.py, which you can use in your own Tornado web applications:

from tornado import web
from projectile.server import TileHandler

...

app = web.Application([
    (r'/([0-9]+)/([0-9]+)/([0-9]+)/([0-9]+).png', TileHandler,
     dict(array=array)),
    ...
])

...

Launching projectile from your own Python code

You can also launch a server from your own Python code with the run() function defined in server.py:

from projectile.server import run

run(array)

Credits

The demo client is lifted from http://bl.ocks.org/mbostock/5914438, with the addition of a small filtering check to prevent the client from requesting tiles which lie beyond the image boundaries.