WZMIAOMIAO/deep-learning-for-image-processing · error · AttributeError

'{}' object has no attribute '{}'

Error message

'{}' object has no attribute '{}'

What it means

This AttributeError is raised by the attribute delegation logic of a metric-tracking class (e.g. SmoothedValue/MetricLogger) in distributed_utils.py. The class routes attribute access first to its `meters` dict, then to instance `__dict__`, and only if both fail does it raise this error. It means you accessed a meter or attribute (e.g. `logger.loss`, `meters.global_avg`) that was never registered or set on the object.

Source

Thrown at pytorch_segmentation/deeplab_v3/train_utils/distributed_utils.py:142

class MetricLogger(object):
    def __init__(self, delimiter="\t"):
        self.meters = defaultdict(SmoothedValue)
        self.delimiter = delimiter

    def update(self, **kwargs):
        for k, v in kwargs.items():
            if isinstance(v, torch.Tensor):
                v = v.item()
            assert isinstance(v, (float, int))
            self.meters[k].update(v)

    def __getattr__(self, attr):
        if attr in self.meters:
            return self.meters[attr]
        if attr in self.__dict__:
            return self.__dict__[attr]
        raise AttributeError("'{}' object has no attribute '{}'".format(
            type(self).__name__, attr))

    def __str__(self):
        loss_str = []
        for name, meter in self.meters.items():
            loss_str.append(
                "{}: {}".format(name, str(meter))
            )
        return self.delimiter.join(loss_str)

    def synchronize_between_processes(self):
        for meter in self.meters.values():
            meter.synchronize_between_processes()

    def add_meter(self, name, meter):
        self.meters[name] = meter

    def log_every(self, iterable, print_freq, header=None):

View on GitHub (pinned to 1ec3fe6f37)

Solutions

  1. Check the exact attribute name in the message and compare against names passed to `metric_logger.update(name=value)` or `create_meters`
  2. Ensure the meter is created/registered in the class `__init__` (e.g. `self.meters.setdefault(name, SmoothedValue())`) before first access
  3. Call `update()` with the attribute before reading it, or use `getattr(logger, attr, default)` for optional metrics
  4. If this fires during unpickling or init, guard against missing keys instead of relying on __getattr__ delegation

Example fix

// before
print(metric_logger.loss.global_avg)  # AttributeError if 'loss' never updated
// after
if hasattr(metric_logger, 'loss'):
    print(metric_logger.loss.global_avg)
# or in the loop:
metric_logger.update(loss=losses.item(), lr=optimizer.param_groups[0]['lr'])
Defensive patterns

Strategy: try-catch

Validate before calling

attrs = set(metric_logger.meters.keys()) | set(metric_logger.__dict__.keys())
assert 'loss' in attrs, f"meter 'loss' not registered; have {attrs}"

Type guard

def has_meter(logger, name):
    return name in logger.meters or name in logger.__dict__

Try / catch

try:
    avg = metric_logger.loss.global_avg
except AttributeError as e:
    logging.warning("missing meter: %s; available: %s", e, list(metric_logger.meters))
    avg = None

Prevention

When it happens

Trigger: Accessing `logger.meters.<name>` or `<logger>.<attr>` before `update(loss=...)` was called with that key; mistyping a meter name (e.g. `train_loss` vs `loss`); accessing `global_avg`/`value`/`avg` style helpers via __getattr__ after instantiation without any update; pickling/unpickling edge cases where __dict__ is empty and __getattr__ recurses or misses.

Common situations: Copy-pasting training loops between projects where the metric names changed; calling `print(logger)`-style code or summary code that assumes standard meters exist; custom loops forgetting to call `metric_logger.update(...)` before reading values; running a modified train loop that logs meters not registered in `__init__` (e.g. `lr`, `grad_norm`).

Related errors

AI-assisted analysis of WZMIAOMIAO/deep-learning-for-image-processing@1ec3fe6f37 (2026-08-30). Data as JSON: /api/errors/5ddab1b5b6a3b95c. Report an issue: GitHub.