CV Color Map

What this is for

A depth map is a grayscale image, and a grayscale depth map is nearly unreadable - you can tell near from far, but you can't tell a 2cm step from a 20cm one without squinting. A false-colour map fixes that: perceptually uniform ramps put the boundaries where your eye can actually find them. Same story for heat maps, confidence maps, and any single-channel score map coming out of a vision pipeline.

This is the standard viewer node for that job, and it's the one to reach for before you start debugging a depth estimator that might be fine.

How it works

It converts the frame to grayscale and applies cv2.applyColorMap with the palette you chose. That's it - the operation is one cv2 call. The reason it exists as a curated node rather than a raw wrapper call is the second half: color_scale.

COLORMAP_VIRIDIS and COLORMAP_INFERNO are perceptually uniform, which is the property that matters when the numbers are data: equal steps in value look like equal steps in colour, so a 0.1 difference looks the same in the dark end as in the bright end. COLORMAP_JET is the old rainbow everybody used in papers for twenty years; it's vivid, it's a lie about ordering, and it makes banding appear where there is none. The author's own tooltip says as much, which is the kind of bluntness you want from a node.

Turn on color_scale and the node appends a vertical colour bar with 0..1 intensity labels beside the result, so the picture carries its own key. It's drawn to stay legible even on tiny inputs - including a 1-pixel image - which matters more than it sounds: a colour-mapped preview is exactly the thing you screenshot into a bug report or a write-up, and a colour ramp with no labels is decoration.

Inputs and outputs

Three inputs, all simple:

• image - a batch is processed frame by frame.
• colormap - the palette dropdown, default COLORMAP_VIRIDIS.
• color_scale - off by default; turn it on when the output leaves ComfyUI.

One output: IMAGE. Feed Save Image for an artefact, Preview Image for a look, or keep chaining - the node outputs an ordinary image, so it composes with anything (blend it over the source at low opacity for a depth-overlay, for instance).

Where it fits

For depth specifically: the pipeline is estimator → CV Color Map → your eyes, and that viewing step is not optional overhead. Depth estimators fail in characteristic ways that are obvious as colour and invisible as grey - a hard seam where the model switched from near-field to far-field interpretation, a smooth ramp where there should be an object boundary, a hole. You'll catch all three in a coloured preview and miss them in greyscale.

It's also the honest alternative to a chart when the "data" really is a picture: use this when the array is an image-shaped thing (depth, saliency, confidence), and the CV Chart * nodes when it's a small table of numbers.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/bmad4ever/comfyui_cv
# restart ComfyUI

Or search ComfyUI CV in ComfyUI Manager (publisher bmad4ever). Requires Python ≥ 3.12, a recent V3-node-API ComfyUI, and opencv-contrib-python-headless~=5.0.0.93 (installed from the pack's requirements). No models, no downloads.

Common issues

• The result is grey-ish and boring. You fed it a colour image and it converted to grayscale first, as documented. This node is a false-colour mapper for single-channel data, not a stylize node.
• Bandit artefacts. That's JET. Switch to viridis or inferno.
• The colour bar is missing. Check color_scale; it's off by default and it's a separate draw step, so it will also cost you a moment on large frames.
• Contrib-backed nodes disappeared pack-wide. A non-contrib OpenCV wheel clobbered the shared site-packages/cv2. Diagnose and repair with python tools/repair_opencv_contrib.py --check / --apply.