How multiscale training data augmentation works on mmdetection? Like, I know images are resized to different resolutions, but how this actually work? Like wasn't the object detector expecting a fixed resolution as input? I tried searching for an answer on blogs and papers, but could not find.
How multiscale training data augmentation works on mmdetection? Like, I know images are resized to different resolutions, but how this actually work? Like wasn't the object detector expecting a fixed resolution as input? I tried searching for an answer on blogs and papers, but could not find.