# -*- coding: utf-8 -*- # Copyright (c) Facebook, Inc. and its affiliates. """ This file contains components with some default boilerplate logic user may need in training / testing. They will not work for everyone, but many users may find them useful. The behavior of functions/classes in this file is subject to change, since they are meant to represent the "common default behavior" people need in their projects. """ import argparse import logging import os import sys import weakref from collections import OrderedDict from typing import Optional import torch from fvcore.nn.precise_bn import get_bn_modules from omegaconf import OmegaConf from torch.nn.parallel import DistributedDataParallel import detectron2.data.transforms as T from detectron2.checkpoint import DetectionCheckpointer from detectron2.config import CfgNode, LazyConfig from detectron2.data import ( MetadataCatalog, build_detection_test_loader, build_detection_train_loader, ) from detectron2.evaluation import ( DatasetEvaluator, inference_on_dataset, print_csv_format, verify_results, ) from detectron2.modeling import build_model from detectron2.solver import build_lr_scheduler, build_optimizer from detectron2.utils import comm from detectron2.utils.collect_env import collect_env_info from detectron2.utils.env import seed_all_rng from detectron2.utils.events import CommonMetricPrinter, JSONWriter, TensorboardXWriter from detectron2.utils.file_io import PathManager from detectron2.utils.logger import setup_logger from . import hooks from .train_loop import AMPTrainer, SimpleTrainer, TrainerBase __all__ = [ "create_ddp_model", "default_argument_parser", "default_setup", "default_writers", "DefaultPredictor", "DefaultTrainer", ] def create_ddp_model(model, *, fp16_compression=False, **kwargs): """ Create a DistributedDataParallel model if there are >1 processes. Args: model: a torch.nn.Module fp16_compression: add fp16 compression hooks to the ddp object. See more at https://pytorch.org/docs/stable/ddp_comm_hooks.html#torch.distributed.algorithms.ddp_comm_hooks.default_hooks.fp16_compress_hook kwargs: other arguments of :module:`torch.nn.parallel.DistributedDataParallel`. """ # noqa if comm.get_world_size() == 1: return model if "device_ids" not in kwargs: kwargs["device_ids"] = [comm.get_local_rank()] ddp = DistributedDataParallel(model, **kwargs) if fp16_compression: from torch.distributed.algorithms.ddp_comm_hooks import default as comm_hooks ddp.register_comm_hook(state=None, hook=comm_hooks.fp16_compress_hook) return ddp def default_argument_parser(epilog=None): """ Create a parser with some common arguments used by detectron2 users. Args: epilog (str): epilog passed to ArgumentParser describing the usage. Returns: argparse.ArgumentParser: """ parser = argparse.ArgumentParser( epilog=epilog or f""" Examples: Run on single machine: $ {sys.argv[0]} --num-gpus 8 --config-file cfg.yaml Change some config options: $ {sys.argv[0]} --config-file cfg.yaml MODEL.WEIGHTS /path/to/weight.pth SOLVER.BASE_LR 0.001 Run on multiple machines: (machine0)$ {sys.argv[0]} --machine-rank 0 --num-machines 2 --dist-url [--other-flags] (machine1)$ {sys.argv[0]} --machine-rank 1 --num-machines 2 --dist-url [--other-flags] """, formatter_class=argparse.RawDescriptionHelpFormatter, ) parser.add_argument("--config-file", default="", metavar="FILE", help="path to config file") parser.add_argument( "--resume", action="store_true", help="Whether to attempt to resume from the checkpoint directory. " "See documentation of `DefaultTrainer.resume_or_load()` for what it means.", ) parser.add_argument("--eval-only", action="store_true", help="perform evaluation only") parser.add_argument("--num-gpus", type=int, default=1, help="number of gpus *per machine*") parser.add_argument("--num-machines", type=int, default=1, help="total number of machines") parser.add_argument( "--machine-rank", type=int, default=0, help="the rank of this machine (unique per machine)" ) # PyTorch still may leave orphan processes in multi-gpu training. # Therefore we use a deterministic way to obtain port, # so that users are aware of orphan processes by seeing the port occupied. port = 2**15 + 2**14 + hash(os.getuid() if sys.platform != "win32" else 1) % 2**14 parser.add_argument( "--dist-url", default="tcp://127.0.0.1:{}".format(port), help="initialization URL for pytorch distributed backend. See " "https://pytorch.org/docs/stable/distributed.html for details.", ) parser.add_argument( "opts", help=""" Modify config options at the end of the command. For Yacs configs, use space-separated "PATH.KEY VALUE" pairs. For python-based LazyConfig, use "path.key=value". """.strip(), default=None, nargs=argparse.REMAINDER, ) return parser def _try_get_key(cfg, *keys, default=None): """ Try select keys from cfg until the first key that exists. Otherwise return default. """ if isinstance(cfg, CfgNode): cfg = OmegaConf.create(cfg.dump()) for k in keys: none = object() p = OmegaConf.select(cfg, k, default=none) if p is not none: return p return default def _highlight(code, filename): try: import pygments except ImportError: return code from pygments.lexers import Python3Lexer, YamlLexer from pygments.formatters import Terminal256Formatter lexer = Python3Lexer() if filename.endswith(".py") else YamlLexer() code = pygments.highlight(code, lexer, Terminal256Formatter(style="monokai")) return code def default_setup(cfg, args): """ Perform some basic common setups at the beginning of a job, including: 1. Set up the detectron2 logger 2. Log basic information about environment, cmdline arguments, and config 3. Backup the config to the output directory Args: cfg (CfgNode or omegaconf.DictConfig): the full config to be used args (argparse.NameSpace): the command line arguments to be logged """ output_dir = _try_get_key(cfg, "OUTPUT_DIR", "output_dir", "train.output_dir") if comm.is_main_process() and output_dir: PathManager.mkdirs(output_dir) rank = comm.get_rank() setup_logger(output_dir, distributed_rank=rank, name="fvcore") logger = setup_logger(output_dir, distributed_rank=rank) logger.info("Rank of current process: {}. World size: {}".format(rank, comm.get_world_size())) logger.info("Environment info:\n" + collect_env_info()) logger.info("Command line arguments: " + str(args)) if hasattr(args, "config_file") and args.config_file != "": logger.info( "Contents of args.config_file={}:\n{}".format( args.config_file, _highlight(PathManager.open(args.config_file, "r").read(), args.config_file), ) ) if comm.is_main_process() and output_dir: # Note: some of our scripts may expect the existence of # config.yaml in output directory path = os.path.join(output_dir, "config.yaml") if isinstance(cfg, CfgNode): logger.info("Running with full config:\n{}".format(_highlight(cfg.dump(), ".yaml"))) with PathManager.open(path, "w") as f: f.write(cfg.dump()) else: LazyConfig.save(cfg, path) logger.info("Full config saved to {}".format(path)) # make sure each worker has a different, yet deterministic seed if specified seed = _try_get_key(cfg, "SEED", "train.seed", default=-1) seed_all_rng(None if seed < 0 else seed + rank) # cudnn benchmark has large overhead. It shouldn't be used considering the small size of # typical validation set. if not (hasattr(args, "eval_only") and args.eval_only): torch.backends.cudnn.benchmark = _try_get_key( cfg, "CUDNN_BENCHMARK", "train.cudnn_benchmark", default=False ) def default_writers(output_dir: str, max_iter: Optional[int] = None): """ Build a list of :class:`EventWriter` to be used. It now consists of a :class:`CommonMetricPrinter`, :class:`TensorboardXWriter` and :class:`JSONWriter`. Args: output_dir: directory to store JSON metrics and tensorboard events max_iter: the total number of iterations Returns: list[EventWriter]: a list of :class:`EventWriter` objects. """ PathManager.mkdirs(output_dir) return [ # It may not always print what you want to see, since it prints "common" metrics only. CommonMetricPrinter(max_iter), JSONWriter(os.path.join(output_dir, "metrics.json")), TensorboardXWriter(output_dir), ] class DefaultPredictor: """ Create a simple end-to-end predictor with the given config that runs on single device for a single input image. Compared to using the model directly, this class does the following additions: 1. Load checkpoint from `cfg.MODEL.WEIGHTS`. 2. Always take BGR image as the input and apply conversion defined by `cfg.INPUT.FORMAT`. 3. Apply resizing defined by `cfg.INPUT.{MIN,MAX}_SIZE_TEST`. 4. Take one input image and produce a single output, instead of a batch. This is meant for simple demo purposes, so it does the above steps automatically. This is not meant for benchmarks or running complicated inference logic. If you'd like to do anything more complicated, please refer to its source code as examples to build and use the model manually. Attributes: metadata (Metadata): the metadata of the underlying dataset, obtained from cfg.DATASETS.TEST. Examples: :: pred = DefaultPredictor(cfg) inputs = cv2.imread("input.jpg") outputs = pred(inputs) """ def __init__(self, cfg): self.cfg = cfg.clone() # cfg can be modified by model self.model = build_model(self.cfg) self.model.eval() if len(cfg.DATASETS.TEST): self.metadata = MetadataCatalog.get(cfg.DATASETS.TEST[0]) checkpointer = DetectionCheckpointer(self.model) checkpointer.load(cfg.MODEL.WEIGHTS) self.aug = T.ResizeShortestEdge( [cfg.INPUT.MIN_SIZE_TEST, cfg.INPUT.MIN_SIZE_TEST], cfg.INPUT.MAX_SIZE_TEST ) self.input_format = cfg.INPUT.FORMAT assert self.input_format in ["RGB", "BGR"], self.input_format def __call__(self, original_image): """ Args: original_image (np.ndarray): an image of shape (H, W, C) (in BGR order). Returns: predictions (dict): the output of the model for one image only. See :doc:`/tutorials/models` for details about the format. """ with torch.no_grad(): # https://github.com/sphinx-doc/sphinx/issues/4258 # Apply pre-processing to image. if self.input_format == "RGB": # whether the model expects BGR inputs or RGB original_image = original_image[:, :, ::-1] height, width = original_image.shape[:2] image = self.aug.get_transform(original_image).apply_image(original_image) image = torch.as_tensor(image.astype("float32").transpose(2, 0, 1)) image.to(self.cfg.MODEL.DEVICE) inputs = {"image": image, "height": height, "width": width} predictions = self.model([inputs])[0] return predictions class DefaultTrainer(TrainerBase): """ A trainer with default training logic. It does the following: 1. Create a :class:`SimpleTrainer` using model, optimizer, dataloader defined by the given config. Create a LR scheduler defined by the config. 2. Load the last checkpoint or `cfg.MODEL.WEIGHTS`, if exists, when `resume_or_load` is called. 3. Register a few common hooks defined by the config. It is created to simplify the **standard model training workflow** and reduce code boilerplate for users who only need the standard training workflow, with standard features. It means this class makes *many assumptions* about your training logic that may easily become invalid in a new research. In fact, any assumptions beyond those made in the :class:`SimpleTrainer` are too much for research. The code of this class has been annotated about restrictive assumptions it makes. When they do not work for you, you're encouraged to: 1. Overwrite methods of this class, OR: 2. Use :class:`SimpleTrainer`, which only does minimal SGD training and nothing else. You can then add your own hooks if needed. OR: 3. Write your own training loop similar to `tools/plain_train_net.py`. See the :doc:`/tutorials/training` tutorials for more details. Note that the behavior of this class, like other functions/classes in this file, is not stable, since it is meant to represent the "common default behavior". It is only guaranteed to work well with the standard models and training workflow in detectron2. To obtain more stable behavior, write your own training logic with other public APIs. Examples: :: trainer = DefaultTrainer(cfg) trainer.resume_or_load() # load last checkpoint or MODEL.WEIGHTS trainer.train() Attributes: scheduler: checkpointer (DetectionCheckpointer): cfg (CfgNode): """ def __init__(self, cfg): """ Args: cfg (CfgNode): """ super().__init__() logger = logging.getLogger("detectron2") if not logger.isEnabledFor(logging.INFO): # setup_logger is not called for d2 setup_logger() cfg = DefaultTrainer.auto_scale_workers(cfg, comm.get_world_size()) # Assume these objects must be constructed in this order. model = self.build_model(cfg) optimizer = self.build_optimizer(cfg, model) data_loader = self.build_train_loader(cfg) model = create_ddp_model(model, broadcast_buffers=False) self._trainer = (AMPTrainer if cfg.SOLVER.AMP.ENABLED else SimpleTrainer)( model, data_loader, optimizer ) self.scheduler = self.build_lr_scheduler(cfg, optimizer) self.checkpointer = DetectionCheckpointer( # Assume you want to save checkpoints together with logs/statistics model, cfg.OUTPUT_DIR, trainer=weakref.proxy(self), ) self.start_iter = 0 self.max_iter = cfg.SOLVER.MAX_ITER self.cfg = cfg self.register_hooks(self.build_hooks()) def resume_or_load(self, resume=True): """ If `resume==True` and `cfg.OUTPUT_DIR` contains the last checkpoint (defined by a `last_checkpoint` file), resume from the file. Resuming means loading all available states (eg. optimizer and scheduler) and update iteration counter from the checkpoint. ``cfg.MODEL.WEIGHTS`` will not be used. Otherwise, this is considered as an independent training. The method will load model weights from the file `cfg.MODEL.WEIGHTS` (but will not load other states) and start from iteration 0. Args: resume (bool): whether to do resume or not """ self.checkpointer.resume_or_load(self.cfg.MODEL.WEIGHTS, resume=resume) if resume and self.checkpointer.has_checkpoint(): # The checkpoint stores the training iteration that just finished, thus we start # at the next iteration self.start_iter = self.iter + 1 def build_hooks(self): """ Build a list of default hooks, including timing, evaluation, checkpointing, lr scheduling, precise BN, writing events. Returns: list[HookBase]: """ cfg = self.cfg.clone() cfg.defrost() cfg.DATALOADER.NUM_WORKERS = 0 # save some memory and time for PreciseBN ret = [ hooks.IterationTimer(), hooks.LRScheduler(), hooks.PreciseBN( # Run at the same freq as (but before) evaluation. cfg.TEST.EVAL_PERIOD, self.model, # Build a new data loader to not affect training self.build_train_loader(cfg), cfg.TEST.PRECISE_BN.NUM_ITER, ) if cfg.TEST.PRECISE_BN.ENABLED and get_bn_modules(self.model) else None, ] # Do PreciseBN before checkpointer, because it updates the model and need to # be saved by checkpointer. # This is not always the best: if checkpointing has a different frequency, # some checkpoints may have more precise statistics than others. if comm.is_main_process(): ret.append(hooks.PeriodicCheckpointer(self.checkpointer, cfg.SOLVER.CHECKPOINT_PERIOD)) def test_and_save_results(): self._last_eval_results = self.test(self.cfg, self.model) return self._last_eval_results # Do evaluation after checkpointer, because then if it fails, # we can use the saved checkpoint to debug. ret.append(hooks.EvalHook(cfg.TEST.EVAL_PERIOD, test_and_save_results)) if comm.is_main_process(): # Here the default print/log frequency of each writer is used. # run writers in the end, so that evaluation metrics are written ret.append(hooks.PeriodicWriter(self.build_writers(), period=20)) return ret def build_writers(self): """ Build a list of writers to be used using :func:`default_writers()`. If you'd like a different list of writers, you can overwrite it in your trainer. Returns: list[EventWriter]: a list of :class:`EventWriter` objects. """ return default_writers(self.cfg.OUTPUT_DIR, self.max_iter) def train(self): """ Run training. Returns: OrderedDict of results, if evaluation is enabled. Otherwise None. """ super().train(self.start_iter, self.max_iter) if len(self.cfg.TEST.EXPECTED_RESULTS) and comm.is_main_process(): assert hasattr( self, "_last_eval_results" ), "No evaluation results obtained during training!" verify_results(self.cfg, self._last_eval_results) return self._last_eval_results def run_step(self): self._trainer.iter = self.iter self._trainer.run_step() def state_dict(self): ret = super().state_dict() ret["_trainer"] = self._trainer.state_dict() return ret def load_state_dict(self, state_dict): super().load_state_dict(state_dict) self._trainer.load_state_dict(state_dict["_trainer"]) @classmethod def build_model(cls, cfg): """ Returns: torch.nn.Module: It now calls :func:`detectron2.modeling.build_model`. Overwrite it if you'd like a different model. """ model = build_model(cfg) logger = logging.getLogger(__name__) logger.info("Model:\n{}".format(model)) return model @classmethod def build_optimizer(cls, cfg, model): """ Returns: torch.optim.Optimizer: It now calls :func:`detectron2.solver.build_optimizer`. Overwrite it if you'd like a different optimizer. """ return build_optimizer(cfg, model) @classmethod def build_lr_scheduler(cls, cfg, optimizer): """ It now calls :func:`detectron2.solver.build_lr_scheduler`. Overwrite it if you'd like a different scheduler. """ return build_lr_scheduler(cfg, optimizer) @classmethod def build_train_loader(cls, cfg): """ Returns: iterable It now calls :func:`detectron2.data.build_detection_train_loader`. Overwrite it if you'd like a different data loader. """ return build_detection_train_loader(cfg) @classmethod def build_test_loader(cls, cfg, dataset_name): """ Returns: iterable It now calls :func:`detectron2.data.build_detection_test_loader`. Overwrite it if you'd like a different data loader. """ return build_detection_test_loader(cfg, dataset_name) @classmethod def build_evaluator(cls, cfg, dataset_name): """ Returns: DatasetEvaluator or None It is not implemented by default. """ raise NotImplementedError( """ If you want DefaultTrainer to automatically run evaluation, please implement `build_evaluator()` in subclasses (see train_net.py for example). Alternatively, you can call evaluation functions yourself (see Colab balloon tutorial for example). """ ) @classmethod def test(cls, cfg, model, evaluators=None): """ Evaluate the given model. The given model is expected to already contain weights to evaluate. Args: cfg (CfgNode): model (nn.Module): evaluators (list[DatasetEvaluator] or None): if None, will call :meth:`build_evaluator`. Otherwise, must have the same length as ``cfg.DATASETS.TEST``. Returns: dict: a dict of result metrics """ logger = logging.getLogger(__name__) if isinstance(evaluators, DatasetEvaluator): evaluators = [evaluators] if evaluators is not None: assert len(cfg.DATASETS.TEST) == len(evaluators), "{} != {}".format( len(cfg.DATASETS.TEST), len(evaluators) ) results = OrderedDict() for idx, dataset_name in enumerate(cfg.DATASETS.TEST): data_loader = cls.build_test_loader(cfg, dataset_name) # When evaluators are passed in as arguments, # implicitly assume that evaluators can be created before data_loader. if evaluators is not None: evaluator = evaluators[idx] else: try: evaluator = cls.build_evaluator(cfg, dataset_name) except NotImplementedError: logger.warn( "No evaluator found. Use `DefaultTrainer.test(evaluators=)`, " "or implement its `build_evaluator` method." ) results[dataset_name] = {} continue results_i = inference_on_dataset(model, data_loader, evaluator) results[dataset_name] = results_i if comm.is_main_process(): assert isinstance( results_i, dict ), "Evaluator must return a dict on the main process. Got {} instead.".format( results_i ) logger.info("Evaluation results for {} in csv format:".format(dataset_name)) print_csv_format(results_i) if len(results) == 1: results = list(results.values())[0] return results @staticmethod def auto_scale_workers(cfg, num_workers: int): """ When the config is defined for certain number of workers (according to ``cfg.SOLVER.REFERENCE_WORLD_SIZE``) that's different from the number of workers currently in use, returns a new cfg where the total batch size is scaled so that the per-GPU batch size stays the same as the original ``IMS_PER_BATCH // REFERENCE_WORLD_SIZE``. Other config options are also scaled accordingly: * training steps and warmup steps are scaled inverse proportionally. * learning rate are scaled proportionally, following :paper:`ImageNet in 1h`. For example, with the original config like the following: .. code-block:: yaml IMS_PER_BATCH: 16 BASE_LR: 0.1 REFERENCE_WORLD_SIZE: 8 MAX_ITER: 5000 STEPS: (4000,) CHECKPOINT_PERIOD: 1000 When this config is used on 16 GPUs instead of the reference number 8, calling this method will return a new config with: .. code-block:: yaml IMS_PER_BATCH: 32 BASE_LR: 0.2 REFERENCE_WORLD_SIZE: 16 MAX_ITER: 2500 STEPS: (2000,) CHECKPOINT_PERIOD: 500 Note that both the original config and this new config can be trained on 16 GPUs. It's up to user whether to enable this feature (by setting ``REFERENCE_WORLD_SIZE``). Returns: CfgNode: a new config. Same as original if ``cfg.SOLVER.REFERENCE_WORLD_SIZE==0``. """ old_world_size = cfg.SOLVER.REFERENCE_WORLD_SIZE if old_world_size == 0 or old_world_size == num_workers: return cfg cfg = cfg.clone() frozen = cfg.is_frozen() cfg.defrost() assert ( cfg.SOLVER.IMS_PER_BATCH % old_world_size == 0 ), "Invalid REFERENCE_WORLD_SIZE in config!" scale = num_workers / old_world_size bs = cfg.SOLVER.IMS_PER_BATCH = int(round(cfg.SOLVER.IMS_PER_BATCH * scale)) lr = cfg.SOLVER.BASE_LR = cfg.SOLVER.BASE_LR * scale max_iter = cfg.SOLVER.MAX_ITER = int(round(cfg.SOLVER.MAX_ITER / scale)) warmup_iter = cfg.SOLVER.WARMUP_ITERS = int(round(cfg.SOLVER.WARMUP_ITERS / scale)) cfg.SOLVER.STEPS = tuple(int(round(s / scale)) for s in cfg.SOLVER.STEPS) cfg.TEST.EVAL_PERIOD = int(round(cfg.TEST.EVAL_PERIOD / scale)) cfg.SOLVER.CHECKPOINT_PERIOD = int(round(cfg.SOLVER.CHECKPOINT_PERIOD / scale)) cfg.SOLVER.REFERENCE_WORLD_SIZE = num_workers # maintain invariant logger = logging.getLogger(__name__) logger.info( f"Auto-scaling the config to batch_size={bs}, learning_rate={lr}, " f"max_iter={max_iter}, warmup={warmup_iter}." ) if frozen: cfg.freeze() return cfg # Access basic attributes from the underlying trainer for _attr in ["model", "data_loader", "optimizer"]: setattr( DefaultTrainer, _attr, property( # getter lambda self, x=_attr: getattr(self._trainer, x), # setter lambda self, value, x=_attr: setattr(self._trainer, x, value), ), )