scripts/train_instance_seg.py [421:442]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    else:
        assert not args.eval, "--resume is needed in eval mode"
        snapshot = None

    # Init GPU stuff
    torch.backends.cudnn.benchmark = config["general"].getboolean("cudnn_benchmark")
    model = DistributedDataParallel(model.cuda(device), device_ids=[device_id], output_device=device_id,
                                    find_unused_parameters=True)

    # Create optimizer
    optimizer, scheduler, batch_update, total_epochs = make_optimizer(config, model, len(train_dataloader))
    if args.resume:
        optimizer.load_state_dict(snapshot["state_dict"]["optimizer"])

    # Training loop
    momentum = 1. - 1. / len(train_dataloader)
    meters = {
        "loss": AverageMeter((), momentum),
        "obj_loss": AverageMeter((), momentum),
        "bbx_loss": AverageMeter((), momentum),
        "roi_cls_loss": AverageMeter((), momentum),
        "roi_bbx_loss": AverageMeter((), momentum),
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



scripts/train_panoptic.py [521:542]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    else:
        assert not args.eval, "--resume is needed in eval mode"
        snapshot = None

    # Init GPU stuff
    torch.backends.cudnn.benchmark = config["general"].getboolean("cudnn_benchmark")
    model = DistributedDataParallel(model.cuda(device), device_ids=[device_id], output_device=device_id,
                                    find_unused_parameters=True)

    # Create optimizer
    optimizer, scheduler, batch_update, total_epochs = make_optimizer(config, model, len(train_dataloader))
    if args.resume:
        optimizer.load_state_dict(snapshot["state_dict"]["optimizer"])

    # Training loop
    momentum = 1. - 1. / len(train_dataloader)
    meters = {
        "loss": AverageMeter((), momentum),
        "obj_loss": AverageMeter((), momentum),
        "bbx_loss": AverageMeter((), momentum),
        "roi_cls_loss": AverageMeter((), momentum),
        "roi_bbx_loss": AverageMeter((), momentum),
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -