Reference for ultralytics/models/utils/loss.py
Note
This file is available at https://github.com/ultralytics/ultralytics/blob/main/ultralytics/models/utils/loss.py. If you spot a problem please help fix it by contributing a Pull Request 🛠️. Thank you 🙏!
ultralytics.models.utils.loss.DETRLoss
DETRLoss(
nc=80,
loss_gain=None,
aux_loss=True,
use_fl=True,
use_vfl=False,
use_uni_match=False,
uni_match_ind=0,
)
Bases: Module
DETR (DEtection TRansformer) Loss class. This class calculates and returns the different loss components for the DETR object detection model. It computes classification loss, bounding box loss, GIoU loss, and optionally auxiliary losses.
Attributes:
Name | Type | Description |
---|---|---|
nc |
int
|
The number of classes. |
loss_gain |
dict
|
Coefficients for different loss components. |
aux_loss |
bool
|
Whether to compute auxiliary losses. |
use_fl |
bool
|
Use FocalLoss or not. |
use_vfl |
bool
|
Use VarifocalLoss or not. |
use_uni_match |
bool
|
Whether to use a fixed layer to assign labels for the auxiliary branch. |
uni_match_ind |
int
|
The fixed indices of a layer to use if |
matcher |
HungarianMatcher
|
Object to compute matching cost and indices. |
fl |
FocalLoss or None
|
Focal Loss object if |
vfl |
VarifocalLoss or None
|
Varifocal Loss object if |
device |
device
|
Device on which tensors are stored. |
Uses default loss_gain if not provided. Initializes HungarianMatcher with preset cost gains. Supports auxiliary losses and various loss types.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
nc
|
int
|
Number of classes. |
80
|
loss_gain
|
dict
|
Coefficients for different loss components. |
None
|
aux_loss
|
bool
|
Use auxiliary losses from each decoder layer. |
True
|
use_fl
|
bool
|
Use FocalLoss. |
True
|
use_vfl
|
bool
|
Use VarifocalLoss. |
False
|
use_uni_match
|
bool
|
Use fixed layer for auxiliary branch label assignment. |
False
|
uni_match_ind
|
int
|
Index of fixed layer for uni_match. |
0
|
Source code in ultralytics/models/utils/loss.py
_get_assigned_bboxes
Assigns predicted bounding boxes to ground truth bounding boxes based on the match indices.
Source code in ultralytics/models/utils/loss.py
_get_index
staticmethod
Returns batch indices, source indices, and destination indices from provided match indices.
Source code in ultralytics/models/utils/loss.py
_get_loss
_get_loss(
pred_bboxes,
pred_scores,
gt_bboxes,
gt_cls,
gt_groups,
masks=None,
gt_mask=None,
postfix="",
match_indices=None,
)
Get losses.
Source code in ultralytics/models/utils/loss.py
_get_loss_aux
_get_loss_aux(
pred_bboxes,
pred_scores,
gt_bboxes,
gt_cls,
gt_groups,
match_indices=None,
postfix="",
masks=None,
gt_mask=None,
)
Get auxiliary losses.
Source code in ultralytics/models/utils/loss.py
_get_loss_bbox
Computes bounding box and GIoU losses for predicted and ground truth bounding boxes.
Source code in ultralytics/models/utils/loss.py
_get_loss_class
Computes the classification loss based on predictions, target values, and ground truth scores.
Source code in ultralytics/models/utils/loss.py
forward
Calculate loss for predicted bounding boxes and scores.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
pred_bboxes
|
Tensor
|
Predicted bounding boxes, shape [l, b, query, 4]. |
required |
pred_scores
|
Tensor
|
Predicted class scores, shape [l, b, query, num_classes]. |
required |
batch
|
dict
|
Batch information containing: cls (torch.Tensor): Ground truth classes, shape [num_gts]. bboxes (torch.Tensor): Ground truth bounding boxes, shape [num_gts, 4]. gt_groups (List[int]): Number of ground truths for each image in the batch. |
required |
postfix
|
str
|
Postfix for loss names. |
''
|
**kwargs
|
Any
|
Additional arguments, may include 'match_indices'. |
{}
|
Returns:
Type | Description |
---|---|
dict
|
Computed losses, including main and auxiliary (if enabled). |
Note
Uses last elements of pred_bboxes and pred_scores for main loss, and the rest for auxiliary losses if self.aux_loss is True.
Source code in ultralytics/models/utils/loss.py
ultralytics.models.utils.loss.RTDETRDetectionLoss
RTDETRDetectionLoss(
nc=80,
loss_gain=None,
aux_loss=True,
use_fl=True,
use_vfl=False,
use_uni_match=False,
uni_match_ind=0,
)
Bases: DETRLoss
Real-Time DeepTracker (RT-DETR) Detection Loss class that extends the DETRLoss.
This class computes the detection loss for the RT-DETR model, which includes the standard detection loss as well as an additional denoising training loss when provided with denoising metadata.
Source code in ultralytics/models/utils/loss.py
forward
Forward pass to compute the detection loss.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
preds
|
tuple
|
Predicted bounding boxes and scores. |
required |
batch
|
dict
|
Batch data containing ground truth information. |
required |
dn_bboxes
|
Tensor
|
Denoising bounding boxes. Default is None. |
None
|
dn_scores
|
Tensor
|
Denoising scores. Default is None. |
None
|
dn_meta
|
dict
|
Metadata for denoising. Default is None. |
None
|
Returns:
Type | Description |
---|---|
dict
|
Dictionary containing the total loss and, if applicable, the denoising loss. |
Source code in ultralytics/models/utils/loss.py
get_dn_match_indices
staticmethod
Get the match indices for denoising.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
dn_pos_idx
|
List[Tensor]
|
List of tensors containing positive indices for denoising. |
required |
dn_num_group
|
int
|
Number of denoising groups. |
required |
gt_groups
|
List[int]
|
List of integers representing the number of ground truths for each image. |
required |
Returns:
Type | Description |
---|---|
List[tuple]
|
List of tuples containing matched indices for denoising. |