YOLO Grayscale Support

This repository extends the functionalities of the ultralytics library by adding support for grayscale images. This functionality has been tested for detection and classification only. Its use for pose and segmentation haven't been tested yet lol.

Installation

To use grayscale image functionality in YOLO, replace the default ultralytics package with the modified version from this repository.

Using CLI

Go to the python packages or anaconda env libraries:

cd /path/to/python/packages

Remove the ultralytics package and clone this repo:

rm -r ultralytics
git clone  git@github.com:Mauricio-Gonzalez-Ortiz/ultralytics.git

Extract the ultralytics folder:

find "ultralytics" -mindepth 1 -path "ultralytics/ultralytics" -prune -o -exec rm -rf {} +
mv ultralytics/ultralytics/* ultralytics/
rm -r ultralytics/ultralytics

Editing your own package

Follow the steps in this medium article

Training

To train a model in python you just need to specify a configuration flag for the number of channels "ch=1"

Python Example

from ultralytics import YOLO

model = YOLO("yolov8.pt")
model.train(data="yaml_file.yaml", ..., ch=1)

CLI Example

yolo detect/classify train data=yaml_file_path/folder_path ... ch=1

Prediction

For prediction, it is recommended to specify the same "ch=1" flag

Python Example

from ultralytics import YOLO
import cv2 

model = YOLO("path/to/your/model.pt")
img = cv2.imread("path/to/your/image", cv2.IMREAD_GRAYSCALE) 
results = model(img, ch=1)

CLI Example

Currently, prediction on grayscale images using the CLI doesn't work.

Exporting in different formats

These changes also allow the user to export a model in a different format in grayscale:

CLI Example

yolo export model="/path/to/your/model.pt" format="engine" ch=1 

If any problems arise when loading a different model type, specify the type of task

from ultralytics import YOLO
import cv2 

model = YOLO("path/to/your/model.engine", task="classify/detect")
img = cv2.imread("path/to/your/image", cv2.IMREAD_GRAYSCALE) 
results = model(img, ch=1)