Potentially 2 LR Schedulers used for Object Detection Tutorial

Link to tutorial: https://pytorch.org/tutorials/intermediate/torchvision_tutorial.html

Hi there, I am following the object detection tutorial and have copied the necessary files from pytorch/vision/main/references/detection/. When inspecting the engine.train_one_epoch() function, I see that it creates a LR scheduler (LinearLR) there and calls lr_scheduler.step() inside train_one_epoch(), however in the tutorial it also creates its own LR scheduler (StepLR) in the main script, and also calls lr_scheduler.step() within the training loop.

Is this meant to be the case or should the tutorial be updated?

cc @datumbox @nairbv @fmassa @NicolasHug @YosuaMichael

Start with the linked object detection tutorial and the copied detection reference files, then trace engine.train_one_epoch() alongside the tutorial's training loop. Compare the LinearLR and StepLR creation and step calls, and confirm the intended scheduler behavior with the project context. Done means the tutorial and reference usage are consistent and the scheduler choice is documented clearly.