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
cfg.DATALOADER.NUM_WORKERS = 0 # save some memory and time for PreciseBN
ret = [
hooks.IterationTimer(),
hooks.LRScheduler(self.optimizer, self.scheduler),
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():
# run writers in the end, so that evaluation metrics are written
ret.append(
hooks.PeriodicWriter(self.build_writers(),
period=cfg.HOOKS.LOG_PERIOD))
return ret
def _write_metrics(self, metrics_dict: dict):
"""
Args:
metrics_dict (dict): dict of scalar metrics
"""
metrics_dict = {
k: v.detach().cpu().item()
if isinstance(v, torch.Tensor) else float(v)
for k, v in metrics_dict.items()
}
# gather metrics among all workers for logging
# This assumes we do DDP-style training, which is currently the only
# supported method in dl_lib.
all_metrics_dict = comm.gather(metrics_dict)
if comm.is_main_process():
if "data_time" in all_metrics_dict[0]:
# data_time among workers can have high variance. The actual latency
# caused by data_time is the maximum among workers.
data_time = np.max(
[x.pop("data_time") for x in all_metrics_dict])
self.storage.put_scalar("data_time", data_time)
# average the rest metrics
metrics_dict = {
k: np.mean([x[k] for x in all_metrics_dict])
for k in all_metrics_dict[0].keys()
}
total_losses_reduced = sum(loss for loss in metrics_dict.values())
self.storage.put_scalar("total_loss", total_losses_reduced)
if len(metrics_dict) > 1:
self.storage.put_scalars(**metrics_dict)
def evaluate(self):
results = OrderedDict()
for evaluator in self._evaluators:
result = evaluator.evaluate()
if is_main_process():
for k, v in result.items():
assert (
k not in results
), "Different evaluators produce results with the same key {}".format(k)
results[k] = v
return results
def __init__(self,
model,
save_dir="",
*,
save_to_disk=None,
**checkpointables):
is_main_process = comm.is_main_process()
super().__init__(
model,
save_dir,
save_to_disk=is_main_process
if save_to_disk is None else save_to_disk,
**checkpointables,
)
def test(cls, cfg, model, evaluators=None):
"""
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
"""
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))
if len(results) == 1:
results = list(results.values())[0]
return results
def evaluate(self):
"""
Returns:
dict: has a key "segm", whose value is a dict of "AP", "AP50", and "AP75".
"""
all_predictions = comm.gather(self._predictions, dst=0)
if not comm.is_main_process():
return
predictions = defaultdict(list)
for predictions_per_rank in all_predictions:
for clsid, lines in predictions_per_rank.items():
predictions[clsid].extend(lines)
del all_predictions
self._logger.info(
"Evaluating {} using {} metric. "
"Note that results do not use the official Matlab API.".format(
self._dataset_name, 2007 if self._is_2007 else 2012
)
)
with tempfile.TemporaryDirectory(prefix="pascal_voc_eval_") as dirname:
res_file_template = os.path.join(dirname, "{}.txt")
aps = defaultdict(list) # iou -> ap per class
for cls_id, cls_name in enumerate(self._class_names):
lines = predictions.get(cls_id, [""])
with open(res_file_template.format(cls_name), "w") as f:
f.write("\n".join(lines))
for thresh in range(50, 100, 5):
rec, prec, ap = voc_eval(
res_file_template,
self._anno_file_template,
self._image_set_path,
cls_name,
ovthresh=thresh / 100.0,
use_07_metric=self._is_2007,
)
aps[thresh].append(ap * 100)
ret = OrderedDict()
mAP = {iou: np.mean(x) for iou, x in aps.items()}
ret["bbox"] = {"AP": np.mean(list(mAP.values())), "AP50": mAP[50], "AP75": mAP[75]}
return ret