ComfyUI Line Segment Detector

A ComfyUI custom node pack that detects line segments in images using the OpenCV LSD (Line Segment Detector) algorithm and provides utilities for visualization, JSON export, and statistics.

Nodes

1. Line Segment Detector (LSD)

Detects line segments from an input image using the LSD algorithm.

Inputs

Name	Type	Default	Description
image	IMAGE	—	Input image (batch supported)
scale	FLOAT	0.8	Internal image scale for LSD. Lower values are faster.
sigma_scale	FLOAT	0.6	Gaussian sigma = sigma_scale / scale
quant	FLOAT	2.0	Bound on gradient norm quantization error
ang_th	FLOAT	22.5	Gradient angle tolerance in degrees
log_eps	FLOAT	0.0	Detection threshold: −log₁₀(NFA) > log_eps
density_th	FLOAT	0.7	Minimum density of aligned points in the enclosing rectangle
min_length	FLOAT	10.0	Discard segments shorter than this value (pixels)

Outputs

Name	Type	Description
line_segments	LINE_SEGMENT	List of detected segments (see data format below)

---

2. Line Segment Visualizer

Draws detected line segments on top of the original image.

Inputs

Name	Type	Default	Description
image	IMAGE	—	Background image to draw on
line_segments	LINE_SEGMENT	—	Segments to draw
color_r	INT	0	Line color — Red channel (0–255)
color_g	INT	255	Line color — Green channel (0–255)
color_b	INT	0	Line color — Blue channel (0–255)
thickness	INT	1	Line thickness in pixels

Outputs: IMAGE

---

3. Line Segment → JSON

Serializes the LINE_SEGMENT data to a JSON string.

Inputs: line_segments (LINE_SEGMENT), indent (INT, default 2)

Outputs: STRING

---

4. Line Segment Stats

Computes summary statistics from detected segments.

Inputs: line_segments (LINE_SEGMENT)

Outputs

Name	Type	Description
count	INT	Number of detected segments
mean_length	FLOAT	Average segment length (px)
max_length	FLOAT	Longest segment length (px)
min_length	FLOAT	Shortest segment length (px)
stats_json	STRING	Full stats as a JSON string (includes std_length)

---

LINE_SEGMENT Data Format

The LINE_SEGMENT type is a plain Python list of dict objects — fully JSON-serializable and easy to process downstream.

[
  {
    "x1": 10.3,
    "y1": 12.7,
    "x2": 118.5,
    "y2": 120.1,
    "length": 152.4,
    "angle": 44.9,
    "width": 1.8,
    "prec": 0.78,
    "nfa": -3.2
  },
  ...
]

Field	Description
x1, y1	Start point (pixels)
x2, y2	End point (pixels)
length	Euclidean length in pixels
angle	Angle in degrees (−180° to 180°, measured from positive x-axis)
width	Estimated line width
prec	Alignment precision
nfa	−log₁₀(NFA) detection score

For batch inputs (more than one image), the output is a list of the above lists — one per image in the batch.

---

Requirements

• opencv-python >= 4.0
• numpy
• torch

No extra model weights are needed — LSD is a classical algorithm built into OpenCV.

pip install opencv-python

---

Installation

Copy or symlink this folder into your ComfyUI custom_nodes directory:

# Option A — copy
cp -r ComfyUI_LineSegmentDetector /path/to/ComfyUI/custom_nodes/

# Option B — symlink
ln -s /absolute/path/to/ComfyUI_LineSegmentDetector \
      /path/to/ComfyUI/custom_nodes/ComfyUI_LineSegmentDetector

Restart ComfyUI. The nodes will appear under the image/analysis category.

---

Example Workflow

Load Image
    │
    ▼
Line Segment Detector (LSD)   ←── tune scale / min_length
    │
    ├──► Line Segment Visualizer ──► Preview Image
    │
    ├──► Line Segment Stats      ──► count / mean_length / …
    │
    └──► Line Segment → JSON     ──► downstream text node

---

License

MIT