training nn is slower then usual almost 2.5 slower

The increased iteration time for your object detection model training could be due to resource contention or changes in the instance's performance. Although you haven't altered the model and only added a small amount of data, variations in CPU utilization, network performance, or overall instance load could be causing slower training. To diagnose the issue, check if GPU utilization is optimal, investigate potential network bottlenecks, and ensure that data loading and preprocessing aren't introducing delays. Also, verify that the instances are performing as expected and consider testing on different instances to rule out hardware variability.