Mask Maker Processor

Overview

The Mask Maker processor automatically finds and segments objects in your image, producing one mask per object and optional Boolean-composite masks. Simply tell it what to look for and where to save the results.

---

Example Usage

- processor:
    processor: "mask_maker"
    inputs:
      source: "input/car_photo.png"
    params:
      cache_json: "cache/car_masks.json" # optional: caches full JSON response
      mask_prompt: "car., window, license plate"
      threshold: 0.2 # detection confidence cutoff
      mask_output: > # compose new masks
        body:  car - window;
        plate: body AND license_plate
    outputs:
      body: "cache/mask_body.png" # <mask> key corresponds to the 'body' mask from mask_output
      plate: "cache/mask_plate.png" # <mask> key corresponds to the 'plate' mask from mask_output
      annotated: "output/annotated_car.png"
    assigns:
      left: "masks_left" # global bounding box key for all masks
      top: "masks_top"
      right: "masks_right"
      bottom: "masks_bottom"
      width: "masks_width"
      height: "masks_height"
      body_left: "body_left" # per-mask bounding box key: '{mask}_left' becomes 'body_left'
      body_top: "body_top"
      plate_left: "plate_left" # per-mask bounding box key: '{mask}_left' becomes 'plate_left'
      plate_top: "plate_top"

---

Parameters

Key	Type	Req’d	Default	Description
cache_json	str	No	—	If provided, caches & reuses the full JSON response
mask_prompt	str	Yes	—	Comma-separated object names; suffix . → keep only the top-scoring instance per term
threshold	float	No	0.2	Confidence cutoff for object detection
mask_output	str	No	identity map	Single-line infix DSL to define composite masks; overrides the one-per-term defaults

**Default **mask_output (when omitted) Produces one mask per prompt term: car:car; window:window; license_plate:license_plate

---

Mask-Output DSL

Compose new masks by combining existing masks using Boolean operators. Use the following syntax:

{new_mask}: {mask_definition}; {new_mask2}: {mask_definition2}; ...

• {mask_definition} can be a composite of existing masks using binary operators and supports the NOT/! unary operator.
• Separate multiple definitions with semicolons (;).

Examples:

mask_output: "rider: person & horse; background: !rider;"

Syntax	Alias	Meaning	Example
A & B	AND	intersection	1 & 0 → 0
A + B or A | B	OR	union	union: a + b
A - B	—	difference (A & !B)	1 - 1 → 0
A ^ B	XOR	exclusive or	1 ^ 1 → 0
!A	NOT	complement	!1 → 0

• Operators and mask names are case-insensitive.
• No parentheses; expressions are evaluated left to right.

Outputs

Key	Required	Description
{mask}	Yes	Path to each generated mask PNG. Replace {mask} with the name defined in mask_output or default terms.
annotated	No	Path to auto-annotated image PNG

---

Assigns

After processing, you’ll receive:

• Global bounding box of all masks combined: left, top, right, bottom, width, height

• Per-mask bounding boxes: {mask}_left, {mask}_top, {mask}_right, {mask}_bottom, {mask}_width, {mask}_height

Replace {mask} with the exact mask name key used in your mask_output (e.g., body_left, plate_width).

---

Just declare:

1. What to detect (mask_prompt)
2. How to compose them (mask_output DSL)
3. Where to save each mask (outputs)