Пример #1
0
def main():
    args = parse_args()
    cfg = Config.fromfile(args.config)
    init_dist(**cfg.dist_params)

    if "stages" in cfg:
        cfg = cfg.stages[args.stage]

    experiment = CIFARExperiment(cfg)

    runner = Runner(
        model=experiment.model,
        optimizers=experiment.optimizers,
        schedulers=experiment.schedulers,
        batch_processor=CIFARBatchProcessor(cfg),
        hooks=experiment.hooks,
        work_dir=experiment.work_dir,
    )

    runner.run(
        data_loaders={
            "train": experiment.dataloader("train"),
            "val": experiment.dataloader("val")
        },
        max_epochs=cfg.max_epochs,
    )
Пример #2
0
def main():
    args = parse_args()
    cfg = Config.fromfile(args.config)
    init_dist(**cfg.dist_params)

    experiment = GANExperiment(cfg)
    runner = Runner(model=experiment.model,
                    optimizers=experiment.optimizers,
                    batch_processor=GANBatchProcessor(cfg),
                    hooks=experiment.hooks,
                    work_dir=experiment.work_dir)

    runner.run(data_loaders={'train': experiment.dataloader('train')},
               max_epochs=cfg.total_epochs,
               resume_from=cfg.resume_from,
               load_from=cfg.load_from)
Пример #3
0
def main():
    args = parse_args()
    cfg = Config.fromfile(args.config)

    if cfg.get("cudnn_benchmark", False):
        torch.backends.cudnn.benchmark = True
    init_dist(**cfg.dist_params)

    experiment = CIFARExperiment(cfg)

    # build data loader
    data_loader = experiment.dataloader("val")

    # build model
    model = experiment.model
    load_checkpoint(model, args.checkpoint, map_location="cpu")

    batch_processor = CIFARBatchProcessor(cfg)

    outputs = multi_gpu_test(model, data_loader, batch_processor)
    outputs = collect_results(outputs, len(data_loader.dataset))
    io.dump(outputs, args.out)
    if dist.is_initialized():
        builder = TileClassifierDDPBuilder(config)
    else:
        builder = TileClassifierDPBuilder(config)

    data_loaders = {x: builder.data_loader(x) for x in config.DATA.keys()}
    batch_processor = TileClassifierBatchProcessor(builder)

    runner = Runner(
        model=builder.model,
        optimizers=builder.optimizers,
        schedulers=builder.schedulers,
        hooks=builder.hooks,
        work_dir=builder.config.WORK_DIR,
        batch_processor=batch_processor,
    )
    runner.run(data_loaders=data_loaders, max_epochs=builder.config.MAX_EPOCHS)


if __name__ == "__main__":
    logging.getLogger().addHandler(logging.StreamHandler())
    args = parse_args()
    config = Config.fromfile(args.config_path)

    if args.is_distributed:
        init_dist(**config.DIST_PARAMS)

    train_func = locals()[config.TRAIN_FUNC]
    train_func(config)