box detection using intelrealsense camera using custom trained model yolov8

barath

i have trained a model to detect and segment the boxes but its not that much accurate
so i have a doubt how to make the model accurate and precise how many data set required for each class
and like do i need to take a single box and take thousang images different angle or random boxes to train
the main objective is to detedt the box and send the co-ordinate to pick and place
i have already trained few to detect only single face so the accuray may increase i thought so but not that much increase

lovro

Hi,

Hard to say exactly without seeing your dataset, but improving accuracy usually comes down to data quality, variation, and precise annotations.

You can check out this example we have for box detection and measurement, which uses a pre-trained model from Roboflow: Box Detection+Measure

Model: https://universe.roboflow.com/are-bla-gpxot/cardboard-box-u35qd/model/1

Might give you a good reference point.