When working with OpenCV should the images be transformed to RGB?

[dfvr1994]

OpenCV loads the images in BGR. When calculating using the functional interface https://github.com/Lightning-AI/torchmetrics/tree/master/src/torchmetrics/functional/image should I first transform my images to RGB?

[Borda]

OpenCV loads images in BGR format, whereas many deep learning frameworks and metrics libraries, including torchmetrics, typically expect images in RGB format. If you use torchmetrics image functions that assume RGB input, you should convert your images from BGR to RGB before passing them in. This is especially important if the metric calculations rely on color information, such as metrics computed on color channels separately.

If your processing or model pipeline already converts images to RGB internally or expects BGR format, then no additional conversion may be needed. But as a general practice, converting OpenCV images from BGR to RGB before using torchmetrics’ RGB-based metric functions is recommended to avoid any channel misinterpretation and ensure accurate metric computation