Cadre

Cadre is a DIY picture frame project. It aims for:

• Colour e-ink display support
• Simple web interface
• Automatic photo updates from your collections
• Easy deployment

It's split up into multiple components:

• Encre: convert image files to a native e-ink display palette
• Affiche: Local web interface
• Expo: Automatic photo updates
• Cru: Image metadata loader

Encre

Convert images to an e-ink display palette.

This is an optional component. If you're not using an e-ink display, you can instead use Affiche with a custom display writer script, see below.

Encre process read a wide variety of image formats (any supported by your libvips install). It performs lightness adjustments and perceptual gamut mapping, and finally dithers to the e-ink display palette. The final byte buffer can be sent to the display hardware.

A command line tool, and both a C++ and a Python API, are available.

Samples results are available in test_data. They were generated for a 7.3" Pimoroni Inky Impression. Keep in mind that current colour e-ink display technology has rather low gamut, so it's normal that the images look washed out. That's what it looks like on the actual display. This tool focuses on accurate colour mapping, it can't do miracles.

Build

• Install pkg-config
• Install Python (3.9 or newer)
• Install libvips (8.15.1 or newer)
• Run cmake --workflow --preset release

Other dependencies are installed by Vcpkg.

You might have to upgrade CMake, Python or your C++ compiler, the logs should tell you.

There are additional notes here specific for Raspberry Pi users.

Running

If the build succeeds, you can use the CLI tool at build/release/cli/encre-cli to test image conversions. For example: build/release/cli/encre-cli test_data/colors.png -p will output test_data/colors.bin (palette'd image as raw unsigned bytes) and test_data/colors_preview.png as a preview. Run with -h for more information.

If you have one of the displays listed below, you can use write_to_display.py to directly write an image to the display. Pass the display name as the first argument.

Supported displays

• Pimoroni Inky Impression:
  • Install requirements-pimoroni_inky
  • Use name "pimoroni_inky"
• Good Display E6 7.3" display (GDEP073E01)
  • Install requirements-GDEP073E01.
  • Use name "GDEP073E01"
• Simulated
  • Use name "simulated"
  • No additional requirements
  • Useful for testing, but does nothing

Other display

If your display isn't in this list, we're open to contributions!

Implement the Display protocol. You can looks at GDEP073E01.py for an example.

To create a custom palette, call py_encre.make_palette_xyz or py_encre.make_palette_lab. If you're lucky, your display data sheet will have the CIE Lab values for each colour, otherwise you can eyeball them... An example is available here.

Options

Options are available to tweak the image processing. Operations are performed in the Oklab perceptual colour space.

• Rotation: apply a rotation (after the EXIF orientation, if applicable)
  • Automatic: landscape is unchanged, portrait is rotated to landscape.
  • Landscape: unchanged
  • Portrait: 90° counter-clockwise
  • Landscape upside-down: 180°
  • Portrait upside-down: 90° clockwise
• Dynamic range: percentage of the original image dynamic range to be rescaled into the output palette. Using 0 will keep the original dynamic range, which will lead to a lot of clipping if the palette has a small dynamic range. Using 1 will force the entirety of the dynamic range into the output, which means no clipping, but lowest contrast possible.
• Exposure: Lightness scale factor (multiply with L component). If not specified, then some basic automatic exposure adjustment is made to bring the image dynamic range into the output palette.
• Brightness: Lightness bias factor (add to L component). If not specified, then some basic automatic brightness adjustment is made to bring the image dynamic range into the output palette.
• Contrast: Slope of the sigmoid function used to compress the image dynamic range into the output palette. Larger values increase contrat in the mid range, at the cost of compressing the shadows and highlights.
• Sharpening: edge sharpening, useful to recover some details after the resize.
• Clipped chroma recovery: α value described here, used to recover some color from the clipped highlights caused by gamut mapping.
• Dither error attenuation: exponential attenuation factor applied to the dither error before diffusion. Helps reduce smearing caused by small errors being diffused over large areas, but comes at the cost of colour accuracy. Values over 1 create a more artistic effect.

Options are also defined in encre_options.json for use by Affiche.

Affiche

Local web interface

After building Encre, create a Python virtual environment, and install the requirements using pip.

Start the server using start.sh. Use stop.sh if you need to stop the server when it's running in the background. You might need to change system settings to make port 80 available without privileges. The simplest solution is to run sudo sysctl -w net.ipv4.ip_unprivileged_port_start=80.

The server should be available at the host's LAN address on port 80.

To post a picture to Affiche without the web interface, simply send it as a multipart/form-data as a file or url key:

curl cadre.local -F [email protected]
curl cadre.local -F url=https://upload.wikimedia.org/wikipedia/commons/7/70/African_leopard_male_%28cropped%29.jpg

If you want image metadata support (EXIF information and geolocation), you need to build Cru.

Configuration

Copy the default config and name it config.json. In this file you can overwrite the following fields:

• TEMP_PATH: Where to write the temporary files, absolute or relative to the server executable
• DISPLAY_WRITER_COMMAND: The command line to run for writing a new image to the display as a list of arguments. As well as the image path, it must accept the --options <json> and --preview <path> arguments to pass in the options and the preview image output path.
• DISPLAY_WRITER_OPTIONS_SCHEMA_PATH: Path to the options schema for the display writer. See encre_options.json for an example.
• DISPLAY_WRITER_OPTIONS: Dictionary of option name and default value override. For example: {"dynamic_range": 0.8}.
• MAP_TILES: URL and options for Leaflet L.tileLayer() constructor. Lets you customize the map shown by Affiche. If you want an English map, I recommend the Thunderforest Atlas tiles (free account required to obtain an API key).
• EXPO_ADDRESS: Hostname and optional port suffix for the Expo server. Must be externally reachable (i.e.: affiche.local instead of localhost). Optional, can be null to disable Expo integration. If empty, then default to the local machine network name.

Tips & Help

You want Affiche (and Expo) to start automatically when the server boots

You can use Cron to launch on boot. Edit the user's crontab using crontab -e and add @reboot cd ~/Cadre/affiche && bash start.sh; cd ~/Cadre/expo && bash start.sh (skip the part after ; if you're not using Expo). You will need to use bash -l if you have environment variables required by Affiche (or Expo) in your bash profile (such as if you followed the Encre Raspberry Pi build instructions).

The connection to your Raspberry Pi is unreliable, especially when using the .local address

Try disabling Wifi power management using iwconfig wlan0 power off. You can make this change persistent on reboot by adding /sbin/iwconfig wlan0 power off to /etc/rc.local.

The Expo link is wrong

More specifically, your local address is cadre.local, and clicking the Expo link opens cadre, which fails to resolve.

Make sure that you have the .local address in the /etc/hosts file, and that it appears before the hostname. For example: 127.0.1.1 cadre.local cadre

Expo

Automatic photo update service

This is an optional component. It's a separate service which maintains a database of photos locations and metadata, and periodically posts one to Affiche.

Create a Python virtual environment, and install the requirements using pip. Additional requirements are also needed for the following features:

• FileSystemCollection: Cru
• AmazonPhotosCollection: requirements-azp

Start the server using start.sh. Use stop.sh if you need to stop the server when it's running in the background.

The server should be available at the host's LAN address on port 21110. You can run Expo on the same host as Affiche or a different one.

Configuration

Copy the default config and name it config.json. In this file you can overwrite the following fields:

• DB_PATH: where to write the Expo persistent data, absolute or relative to the server executable

Collections

List all collections by GETing from /collections. Create a collection by PUTting to /collections a JSON object like so:

{
    "identifier": "my_collection",
    "display_name": "My Collection",
    "schedule": "*/5 * * * *",
    "enabled": true,
    "class_name": "FileSystemCollection",
    "settings": {
        // See below
    }
}

• identifier must be unique, contain only characters in the set [A-Za-z0-9_], and cannot start with a number
• display_name is optional and defaults to the identifier value
• schedule Cron expression, or empty string to never run automatically
• enabled is optional and defaults to true
• class_name can only be FileSystemCollection
• settings depends on the class_name value, see below

You can edit a collection by PATCHing to /collections?identifier=<identifier> using the same JSON format, except all fields are now optional. You can also query using GET, and delete using DELETE.

FileSystemCollection

Use this to display photos from a local collection such as a NAS. You can also create an SMB share on a Raspberry Pi where you upload your favourite photos. The photos must be visible on the filesystem where Expo is running.

Requires Cru to process images. Scan the root_path for known image formats.

Does not support the favorite filter.

Settings:

{
    "root_path": "<path to your local photos folder>"
}

AmazonPhotosCollection

Uses an unofficial Amazon Photos Python API by Trevor Hobenshield (actually a fork with a few improvements).

Settings:

{
    "user_agent": "<User agent string from the browser used to login to Amazon Photos>",
    "cookies": {
        // Copy cookies here as a dictionary. It's not known exactly which cookies are required,
        // if you're missing some you might eventually get an authentication error.
        // This list just an example, since the cookie names are region specific (amazon.ca shown).
        // See: https://github.com/trevorhobenshield/amazon_photos?tab=readme-ov-file#setup
        "at-acbca": "***",
        "ubid-acbca": "***",
        "sess-at-acbca": "***",
        "sst-acbca": "***",
        "x-acbca": "***",
        "lc-acbca": "***",
        "session-id": "***",
        "session-id-time": "***"
    }
}

Scan

Immediately trigger a collection scan by POSTing to /scan a JSON object like so:

{
    "identifier": "<collection identifier>",
    "delay": 0
}

• identifier is a collection identifier
• delay a delay in seconds (float), is optional and defaults to 0

Schedules

The schedule API uses the same methods as collections, but on the /schedules endpoint. There is a new optional hostname argument for GET queries, to filter schedules by the target hostname. It will be compared against the original hostname and the external one if they're different (ex.: localhost and affiche.local).

The JSON format is:

{
    "identifier": "my_schedule",
    "display_name": "My Schedule",
    "hostname": "<Affiche hostname>",
    "schedule": "*/20 * * * *",
    "enabled": true,
    "filter": "<filter expression>",
    "order": "<order enum name>",
    "affiche_options": {
        // Example for Encre
        "contrast": 0.5,
        "sharpening": 2
    }
}

• identifier must be unique, contain only characters in the set [A-Za-z0-9_], and cannot start with a number
• display_name is optional and defaults to the identifier value
• hostname is the hostname (optionally with a :<port> suffix) where an Affiche instance is running
• schedule Cron expression, or empty string to never run automatically
• enabled is optional and defaults to true
• filter is optional and defaults to "true"
• order is optional and defaults to SHUFFLE, see below
• affiche_options is optional, override the default options for the target Affiche instance

Filter expressions

The filter EBNF grammar is:

bool literal = "false" | "true";
identifier = ? /[A-Za-z_][A-Za-z0-9_]*/ ?;
favorite = "favorite";
aspect = "landscape" | "portrait" | "square";
collection set = "{", identifier, {identifier}, "}";
parenthesized expression = "(", expression, ")";
atom = bool literal | favorite | aspect | collection set | parenthesized expression;

unary = ["not"], atom;
and = unary, ["and", unary];
or = and, ["or", and];

expression = or;

In simple terms this means you can use the favorite (unused), landscape, portrait and square predicates to write a logical expression for filtering photos. You can use the (), not, and, and or operators to build your expression (listed in decreasing order of precedence). To filter by collection, use {<identifier 1> <identifier 2> ...}. A photo will only be picked if it belongs to one of the given collections. The true and false literals are also available.

Example: portrait and (favorite or not {phone_pics family_pics})

If you simply want to allow all photos use: true

Order options

• SHUFFLE: Cycle through images in a random order. After all images have been displayed once, restart in a different random order.
• CHRONOLOGICAL_DESCENDING and CHRONOLOGICAL_ASCENDING: Cycle through images in descending or ascending capture date. After all images have been displayed once, restart the cycle. Images without a capture date in their metadata are ignored.

Refresh

Immediately trigger a schedule by POSTing to /refresh a JSON object like so:

{
    "identifier": "<schedule identifier>",
    "delay": 0
}

• identifier is a schedule identifier
• delay a delay in seconds (float), is optional and defaults to 0

Cru

Cru is a native Python module for quickly loading image metadata, like resolution and tags. It also does some processing and pretty formatting for easier consumption.

Build

• Follow the Encre build instructions first, that'll make sure you have everything installed
• Run cmake --workflow --preset release

Running

The filesystem collection should automatically detect and import the module. If you see "Cru module not found" in the logs, check that the module can be indeed be loaded from <repo path>/cru/build/release by Python.