Question about Input Image Format [RGB or BGR]

[meherajj]

Hi,

I have a question regarding the input image format used to train the models in this repository. Specifically, I would like to know whether the models were trained on RGB or BGR images.

There are two pretrained models provided: resnet50_final and mobilenet0.25_final. Were these models trained on images in the BGR format?

The reason I ask is that in the detect.py script, images are read using OpenCV and then directly passed to the model. As far as I know, OpenCV reads images in BGR format by default.

Additionally, would the model performance degrade if the images are given in the wrong format (RGB instead of BGR or vice versa)?

Thank you for your help!