5fa1a76
1
2
Important to note is that the visual backbone expects BGR channels instead of RGB, as all models in Detectron2 are pre-trained using the BGR format.
