Verify and adjust data handling in 'train_step'

[clementpoiret]

Review and update the train_step function in hackathon/objectdetection.py to ensure it correctly unpacks and uses batch data (images, objects/bounding boxes, labels/classes) as prepared by the updated data pipeline and expected by the new loss function.

[bernardev254]

With experience in machine learning pipelines and model training, I will review and update the train_step function in hackathon/objectdetection.py. This update will ensure the function correctly unpacks and utilizes batch data, including images, bounding boxes, and labels, as prepared by the updated data pipeline and required by the new loss function.