realtime package¶

The realtime package provides modules for fast, low-latency execution of the VISWIR pipeline. It is designed for scenarios where image fusion and detection must be performed in near real-time, such as embedded systems or live video processing.

Warning

The fast mode does not compute quality metrics. It is mainly intended for quickly testing parameters and obtaining a visual result without the overhead of full evaluation.

Submodules¶

realtime.fast_config module¶

Configuration utilities optimized for fast execution.

Configuration loader and classes for the fast real-time VISWIR pipeline.

realtime.fast_config.load_fast_config(project_root: Path | None = None) → dict[source]¶

Load the minimal configuration for the fast pipeline.

This function loads:

Base fusion parameters from config/fast_config.yaml.
YOLO detection parameters from config/yolo_config.json if available, otherwise falls back to default YOLO settings.

Parameters:

project_root (Path or None, optional) – Root directory of the project. If None, it is inferred automatically by traversing up from the current file location.

Returns:

Dictionary containing the following keys:

base (dict) – Base fusion parameters loaded from YAML.
yolo (dict) – YOLO detection parameters (from JSON or defaults).
paths (dict) – Paths used in the configuration:
- project_root (str)
- config_dir (str)

Return type:

dict

Notes

If yolo_config.json does not exist, default YOLO parameters are used: model path, confidence threshold, IoU threshold, device, and allowed classes.
Ensures consistent configuration loading for both fusion and detection.

realtime.fast_detection module¶

Lightweight detection routines adapted for speed in test contexts.

Real-time object detection module for the fast pipeline.

class realtime.fast_detection.FastDetector(model_path: Path, conf_thres=0.25, iou_thres=0.3, device='cpu', allowed_classes: Sequence[str] | None = None)[source]¶

Bases: object

Fast object detector using Ultralytics YOLO.

This class wraps a YOLO model for quick inference and visualization. It supports filtering by allowed classes and drawing bounding boxes with labels on images.

Parameters:

model_path (Path) – Path to the YOLO model weights.
conf_thres (float, default=0.25) – Confidence threshold for predictions.
iou_thres (float, default=0.3) – IoU threshold for non-maximum suppression.
device (str, default="cpu") – Device to run inference on (“cpu” or “cuda”).
allowed_classes (Sequence[str], optional) – List of class names to allow. If None, all classes are allowed.

Raises:

RuntimeError – If Ultralytics YOLO is not installed.
FileNotFoundError – If the YOLO weights file does not exist.

draw(img_rgb: ndarray, results) → ndarray[source]¶

Draw bounding boxes and labels on an image based on YOLO results.

Parameters:

img_rgb (numpy.ndarray) – Input RGB image (float in [0, 1]).
results (list) – YOLO detection results containing bounding boxes and class IDs.

Returns:

Image with bounding boxes and labels drawn (float in [0, 1]).

Return type:

numpy.ndarray

Notes

Bounding boxes are drawn using skimage.draw.rectangle_perimeter.
Labels are drawn using Pillow.
Colors are assigned per class (e.g., person=green, others=red).

static draw_label(img_rgb: ndarray, x1, y1, name, color=(0, 255, 0))[source]¶

Draw a text label on an image at the given coordinates. (Static method)

Parameters:

img_rgb (numpy.ndarray) – Input RGB image (float in [0, 1]).
x1 (int) – X-coordinate of the label position.
y1 (int) – Y-coordinate of the label position.
name (str) – Class name to display.
color (tuple of int, default=(0, 255, 0)) – RGB color of the text.

Returns:

Image with the label drawn (float in [0, 1]).

Return type:

numpy.ndarray

predict(img_rgb: ndarray)[source]¶

Run YOLO prediction on an RGB image.

Parameters:: img_rgb (numpy.ndarray) – Input image in RGB format, normalized to [0, 1] or [0, 255].
Returns:: List of YOLO detection results.
Return type:: list

realtime.fast_fusion_runner module¶

Runner for executing the fusion pipeline, coordinating fast configuration and detection.

Fast/real-time pipeline runner for quick VISWIR execution.

class realtime.fast_fusion_runner.FastFusionPipeline(cfg: dict)[source]¶

Bases: object

Fast pipeline for VISWIR image fusion with optional YOLO detection.

This class loads configuration parameters, initializes the YOLO detector if required, and provides a run method to fuse visible and SWIR images.

Parameters:

cfg (dict) –

Configuration dictionary containing:

base (dict) – Base fusion parameters (facteur_swir, beta, level, apply_gamma, gamma_value).
yolo (dict) – YOLO detection parameters (model_path, thresholds, device, allowed_classes).

cfg¶

Full configuration dictionary.

Type:: dict

detector¶

YOLO detector instance if detection is enabled, otherwise None.

Type:: FastDetector or None

run_detection¶

Whether detection is enabled (from config).

Type:: bool

defaults¶

Default fusion parameters loaded from config.

Type:: dict

run(visible_path: Path, swir_path: Path, override_params: dict | None = None) → ndarray[source]¶

Run the fusion pipeline on a pair of visible and SWIR images.

Parameters:

visible_path (Path) – Path to the visible (RGB) image.
swir_path (Path) – Path to the SWIR (grayscale) image.
override_params (dict, optional) – Dictionary of parameters to override defaults.

Returns:

Fused RGB image as float64 normalized to [0, 1]. If detection is enabled, bounding boxes and labels are drawn.

Return type:

numpy.ndarray

realtime.fast_fusion_runner.read_image_swir(path: Path) → ndarray[source]¶

Read a SWIR (Short-Wave Infrared) image and convert it to float64 in [0, 1].

Parameters:: path (Path) – Path to the SWIR image file.
Returns:: Grayscale image as float64 normalized to [0, 1].
Return type:: numpy.ndarray

realtime.fast_fusion_runner.read_image_visible(path: Path) → ndarray[source]¶

Read a visible (RGB) image and convert it to float64 in [0, 1].

Parameters:: path (Path) – Path to the visible image file.
Returns:: RGB image as float64 normalized to [0, 1].
Return type:: numpy.ndarray

Module contents¶

The top-level realtime module re-exports selected functions and classes from its submodules for convenience.

realtime package¶

Modules for the fast VISWIR pipeline: - Minimal configuration loading - Fast detection (YOLO) - Fast fusion runner