Generalized Intersection Over Union (gIoU)

Module Interface

class torchmetrics.detection.giou.GeneralizedIntersectionOverUnion(box_format='xyxy', iou_threshold=None, class_metrics=False, respect_labels=True, **kwargs)

Compute Generalized Intersection Over Union (GIoU).

As input to forward and update the metric accepts the following input:

• preds (List): A list consisting of dictionaries each containing the key-values (each dictionary corresponds to a single image). Parameters that should be provided per dict:

  • boxes (Tensor): float tensor of shape (num_boxes, 4) containing num_boxes detection boxes of the format specified in the constructor. By default, this method expects (xmin, ymin, xmax, ymax) in absolute image coordinates.

  • labels (Tensor): integer tensor of shape (num_boxes) containing 0-indexed detection classes for the boxes.

• target (List): A list consisting of dictionaries each containing the key-values (each dictionary corresponds to a single image). Parameters that should be provided per dict:

  • boxes (Tensor): float tensor of shape (num_boxes, 4) containing num_boxes ground truth boxes of the format specified in the constructor. By default, this method expects (xmin, ymin, xmax, ymax) in absolute image coordinates.

  • labels (Tensor): integer tensor of shape (num_boxes) containing 0-indexed ground truth classes for the boxes.

As output of forward and compute the metric returns the following output:

• giou_dict: A dictionary containing the following key-values:

  • giou: (Tensor) with overall giou value over all classes and samples.

  • giou/cl_{cl}: (Tensor), if argument class metrics=True

Parameters:

• box_format (str) – Input format of given boxes. Supported formats are [`xyxy`, `xywh`, `cxcywh`].

• iou_thresholds – Optional IoU thresholds for evaluation. If set to None the threshold is ignored.

• class_metrics (bool) – Option to enable per-class metrics for IoU. Has a performance impact.

• respect_labels (bool) – Ignore values from boxes that do not have the same label as the ground truth box. Else will compute Iou between all pairs of boxes.

• kwargs (Any) – Additional keyword arguments, see Advanced metric settings for more info.

Example

>>> import torch
>>> from torchmetrics.detection import GeneralizedIntersectionOverUnion
>>> preds = [
...    {
...        "boxes": torch.tensor([[296.55, 93.96, 314.97, 152.79], [298.55, 98.96, 314.97, 151.79]]),
...        "scores": torch.tensor([0.236, 0.56]),
...        "labels": torch.tensor([4, 5]),
...    }
... ]
>>> target = [
...    {
...        "boxes": torch.tensor([[300.00, 100.00, 315.00, 150.00]]),
...        "labels": torch.tensor([5]),
...    }
... ]
>>> metric = GeneralizedIntersectionOverUnion()
>>> metric(preds, target)
{'giou': tensor(0.8613)}

Raises:

ModuleNotFoundError – If torchvision is not installed with version 0.8.0 or newer.

plot(val=None, ax=None)

Plot a single or multiple values from the metric.

Parameters:

• val (Union[Tensor, Sequence[Tensor], None]) – Either a single result from calling metric.forward or metric.compute or a list of these results. If no value is provided, will automatically call metric.compute and plot that result.

• ax (Optional[Axes]) – An matplotlib axis object. If provided will add plot to that axis

Return type:

Tuple[Figure, Union[Axes, ndarray]]

Returns:

Figure object and Axes object

Raises:

ModuleNotFoundError – If matplotlib is not installed

>>> # Example plotting single value
>>> import torch
>>> from torchmetrics.detection import GeneralizedIntersectionOverUnion
>>> preds = [
...    {
...        "boxes": torch.tensor([[296.55, 93.96, 314.97, 152.79], [298.55, 98.96, 314.97, 151.79]]),
...        "scores": torch.tensor([0.236, 0.56]),
...        "labels": torch.tensor([4, 5]),
...    }
... ]
>>> target = [
...    {
...        "boxes": torch.tensor([[300.00, 100.00, 315.00, 150.00]]),
...        "labels": torch.tensor([5]),
...    }
... ]
>>> metric = GeneralizedIntersectionOverUnion()
>>> metric.update(preds, target)
>>> fig_, ax_ = metric.plot()

>>> # Example plotting multiple values
>>> import torch
>>> from torchmetrics.detection import GeneralizedIntersectionOverUnion
>>> preds = [
...    {
...        "boxes": torch.tensor([[296.55, 93.96, 314.97, 152.79], [298.55, 98.96, 314.97, 151.79]]),
...        "scores": torch.tensor([0.236, 0.56]),
...        "labels": torch.tensor([4, 5]),
...    }
... ]
>>> target = lambda : [
...    {
...        "boxes": torch.tensor([[300.00, 100.00, 335.00, 150.00]]) + torch.randint(-10, 10, (1, 4)),
...        "labels": torch.tensor([5]),
...    }
... ]
>>> metric = GeneralizedIntersectionOverUnion()
>>> vals = []
>>> for _ in range(20):
...     vals.append(metric(preds, target()))
>>> fig_, ax_ = metric.plot(vals)

Functional Interface

torchmetrics.functional.detection.giou.generalized_intersection_over_union(preds, target, iou_threshold=None, replacement_val=0, aggregate=True)

Compute Generalized Intersection over Union (GIOU) between two sets of boxes.

Both sets of boxes are expected to be in (x1, y1, x2, y2) format with 0 <= x1 < x2 and 0 <= y1 < y2.

Parameters:

• preds (Tensor) – The input tensor containing the predicted bounding boxes.

• target (Tensor) – The tensor containing the ground truth.

• iou_threshold (Optional[float]) – Optional IoU thresholds for evaluation. If set to None the threshold is ignored.

• replacement_val (float) – Value to replace values under the threshold with.

• aggregate (bool) – Return the average value instead of the full matrix of values

Return type:

Tensor

Example::

By default giou is aggregated across all box pairs e.g. mean along the diagonal of the gIoU matrix:

>>> import torch
>>> from torchmetrics.functional.detection import generalized_intersection_over_union
>>> preds = torch.tensor(
...     [
...         [296.55, 93.96, 314.97, 152.79],
...         [328.94, 97.05, 342.49, 122.98],
...         [356.62, 95.47, 372.33, 147.55],
...     ]
... )
>>> target = torch.tensor(
...     [
...         [300.00, 100.00, 315.00, 150.00],
...         [330.00, 100.00, 350.00, 125.00],
...         [350.00, 100.00, 375.00, 150.00],
...     ]
... )
>>> generalized_intersection_over_union(preds, target)
tensor(0.5638)

Example::

By setting aggregate=False the full IoU matrix is returned:

>>> import torch
>>> from torchmetrics.functional.detection import generalized_intersection_over_union
>>> preds = torch.tensor(
...     [
...         [296.55, 93.96, 314.97, 152.79],
...         [328.94, 97.05, 342.49, 122.98],
...         [356.62, 95.47, 372.33, 147.55],
...     ]
... )
>>> target = torch.tensor(
...     [
...         [300.00, 100.00, 315.00, 150.00],
...         [330.00, 100.00, 350.00, 125.00],
...         [350.00, 100.00, 375.00, 150.00],
...     ]
... )
>>> generalized_intersection_over_union(preds, target, aggregate=False)
tensor([[ 0.6895, -0.4964, -0.4944],
        [-0.5105,  0.4673, -0.3434],
        [-0.6024, -0.4021,  0.5345]])