I am currently working with 5x5cm resolution, and made some runs with 1x1cm even. I abandoned the object detection because i didn't fit my goal, so I can't help you with that...
But I made a progress with pixel classification though: I just started again (exporting training data and training the model) and now it worked -_- only thing i altered was that I didn't clip the raster before the workflow this time. Maybe Clipping could be a problem? But sounds very unlikely...