Strip dfl_loss from BboxLoss (#14041)

Co-authored-by: UltralyticsAssistant <web@ultralytics.com>
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>

docs/en/guides/model-evaluation-insights.md

@@ -8,9 +8,9 @@ keywords: Model Evaluation, Machine Learning Model Evaluation, Fine Tuning Machi
 
 ## Introduction
 
-Once you’ve [trained](./model-training-tips.md) your computer vision model, evaluating and refining it to perform optimally is essential. Just training your model isn’t enough. You need to make sure that your model is accurate, efficient, and fulfills the [objective](./defining-project-goals.md) of your computer vision project. By evaluating and fine-tuning your model, you can identify weaknesses, improve its accuracy, and boost overall performance.
+Once you've [trained](./model-training-tips.md) your computer vision model, evaluating and refining it to perform optimally is essential. Just training your model isn't enough. You need to make sure that your model is accurate, efficient, and fulfills the [objective](./defining-project-goals.md) of your computer vision project. By evaluating and fine-tuning your model, you can identify weaknesses, improve its accuracy, and boost overall performance.
 
-In this guide, we’ll share insights on model evaluation and fine-tuning that’ll make this [step of a computer vision project](./steps-of-a-cv-project.md) more approachable. We’ll discuss how to understand evaluation metrics and implement fine-tuning techniques, giving you the knowledge to elevate your model's capabilities.
+In this guide, we'll share insights on model evaluation and fine-tuning that'll make this [step of a computer vision project](./steps-of-a-cv-project.md) more approachable. We'll discuss how to understand evaluation metrics and implement fine-tuning techniques, giving you the knowledge to elevate your model's capabilities.
 
 ## Evaluating Model Performance Using Metrics
 
@@ -34,7 +34,7 @@ Intersection over Union (IoU) is a metric in object detection that measures how
 
 Mean Average Precision (mAP) is a way to measure how well an object detection model performs. It looks at the precision of detecting each object class, averages these scores, and gives an overall number that shows how accurately the model can identify and classify objects.
 
-Let’s focus on two specific mAP metrics:
+Let's focus on two specific mAP metrics:
 
 - *mAP@.5:* Measures the average precision at a single IoU (Intersection over Union) threshold of 0.5. This metric checks if the model can correctly find objects with a looser accuracy requirement. It focuses on whether the object is roughly in the right place, not needing perfect placement. It helps see if the model is generally good at spotting objects.
 - *mAP@.5:.95:* Averages the mAP values calculated at multiple IoU thresholds, from 0.5 to 0.95 in 0.05 increments. This metric is more detailed and strict. It gives a fuller picture of how accurately the model can find objects at different levels of strictness and is especially useful for applications that need precise object detection.
@@ -136,4 +136,4 @@ Sharing your ideas and questions with other computer vision enthusiasts can insp
 
 ## Final Thoughts
 
-Evaluating and fine-tuning your computer vision model are important steps for successful model deployment. These steps help make sure that your model is accurate, efficient, and suited to your overall application. The key to training the best model possible is continuous experimentation and learning. Don’t hesitate to tweak parameters, try new techniques, and explore different datasets. Keep experimenting and pushing the boundaries of what's possible!
+Evaluating and fine-tuning your computer vision model are important steps for successful model deployment. These steps help make sure that your model is accurate, efficient, and suited to your overall application. The key to training the best model possible is continuous experimentation and learning. Don't hesitate to tweak parameters, try new techniques, and explore different datasets. Keep experimenting and pushing the boundaries of what's possible!

docs/en/reference/utils/loss.md

@@ -19,6 +19,10 @@ keywords: Ultralytics, loss functions, Varifocal Loss, Focal Loss, Bbox Loss, Ro
 
 <br><br>
 
+## ::: ultralytics.utils.loss.DFLoss
+
+<br><br>
+
 ## ::: ultralytics.utils.loss.BboxLoss
 
 <br><br>

ultralytics/utils/loss.py

@@ -61,39 +61,22 @@ class FocalLoss(nn.Module):
         return loss.mean(1).sum()
 
 
-class BboxLoss(nn.Module):
+class DFLoss(nn.Module):
-    """Criterion class for computing training losses during training."""
+    """Criterion class for computing DFL losses during training."""
 
-    def __init__(self, reg_max, use_dfl=False):
+    def __init__(self, reg_max=16) -> None:
-        """Initialize the BboxLoss module with regularization maximum and DFL settings."""
+        """Initialize the DFL module."""
         super().__init__()
         self.reg_max = reg_max
-        self.use_dfl = use_dfl
 
-    def forward(self, pred_dist, pred_bboxes, anchor_points, target_bboxes, target_scores, target_scores_sum, fg_mask):
+    def __call__(self, pred_dist, target):
-        """IoU loss."""
-        weight = target_scores.sum(-1)[fg_mask].unsqueeze(-1)
-        iou = bbox_iou(pred_bboxes[fg_mask], target_bboxes[fg_mask], xywh=False, CIoU=True)
-        loss_iou = ((1.0 - iou) * weight).sum() / target_scores_sum
-
-        # DFL loss
-        if self.use_dfl:
-            target_ltrb = bbox2dist(anchor_points, target_bboxes, self.reg_max)
-            loss_dfl = self._df_loss(pred_dist[fg_mask].view(-1, self.reg_max + 1), target_ltrb[fg_mask]) * weight
-            loss_dfl = loss_dfl.sum() / target_scores_sum
-        else:
-            loss_dfl = torch.tensor(0.0).to(pred_dist.device)
-
-        return loss_iou, loss_dfl
-
-    @staticmethod
-    def _df_loss(pred_dist, target):
         """
         Return sum of left and right DFL losses.
 
         Distribution Focal Loss (DFL) proposed in Generalized Focal Loss
         https://ieeexplore.ieee.org/document/9792391
         """
+        target = target.clamp_(0, self.reg_max - 1 - 0.01)
         tl = target.long()  # target left
         tr = tl + 1  # target right
         wl = tr - target  # weight left
@@ -104,12 +87,37 @@ class BboxLoss(nn.Module):
         ).mean(-1, keepdim=True)
 
 
+class BboxLoss(nn.Module):
+    """Criterion class for computing training losses during training."""
+
+    def __init__(self, reg_max=16):
+        """Initialize the BboxLoss module with regularization maximum and DFL settings."""
+        super().__init__()
+        self.dfl_loss = DFLoss(reg_max) if reg_max > 1 else None
+
+    def forward(self, pred_dist, pred_bboxes, anchor_points, target_bboxes, target_scores, target_scores_sum, fg_mask):
+        """IoU loss."""
+        weight = target_scores.sum(-1)[fg_mask].unsqueeze(-1)
+        iou = bbox_iou(pred_bboxes[fg_mask], target_bboxes[fg_mask], xywh=False, CIoU=True)
+        loss_iou = ((1.0 - iou) * weight).sum() / target_scores_sum
+
+        # DFL loss
+        if self.dfl_loss:
+            target_ltrb = bbox2dist(anchor_points, target_bboxes, self.dfl_loss.reg_max - 1)
+            loss_dfl = self.dfl_loss(pred_dist[fg_mask].view(-1, self.dfl_loss.reg_max), target_ltrb[fg_mask]) * weight
+            loss_dfl = loss_dfl.sum() / target_scores_sum
+        else:
+            loss_dfl = torch.tensor(0.0).to(pred_dist.device)
+
+        return loss_iou, loss_dfl
+
+
 class RotatedBboxLoss(BboxLoss):
     """Criterion class for computing training losses during training."""
 
-    def __init__(self, reg_max, use_dfl=False):
+    def __init__(self, reg_max):
         """Initialize the BboxLoss module with regularization maximum and DFL settings."""
-        super().__init__(reg_max, use_dfl)
+        super().__init__(reg_max)
 
     def forward(self, pred_dist, pred_bboxes, anchor_points, target_bboxes, target_scores, target_scores_sum, fg_mask):
         """IoU loss."""
@@ -118,9 +126,9 @@ class RotatedBboxLoss(BboxLoss):
         loss_iou = ((1.0 - iou) * weight).sum() / target_scores_sum
 
         # DFL loss
-        if self.use_dfl:
+        if self.dfl_loss:
-            target_ltrb = bbox2dist(anchor_points, xywh2xyxy(target_bboxes[..., :4]), self.reg_max)
+            target_ltrb = bbox2dist(anchor_points, xywh2xyxy(target_bboxes[..., :4]), self.dfl_loss.reg_max - 1)
-            loss_dfl = self._df_loss(pred_dist[fg_mask].view(-1, self.reg_max + 1), target_ltrb[fg_mask]) * weight
+            loss_dfl = self.dfl_loss(pred_dist[fg_mask].view(-1, self.dfl_loss.reg_max), target_ltrb[fg_mask]) * weight
             loss_dfl = loss_dfl.sum() / target_scores_sum
         else:
             loss_dfl = torch.tensor(0.0).to(pred_dist.device)
@@ -165,18 +173,19 @@ class v8DetectionLoss:
         self.use_dfl = m.reg_max > 1
 
         self.assigner = TaskAlignedAssigner(topk=tal_topk, num_classes=self.nc, alpha=0.5, beta=6.0)
-        self.bbox_loss = BboxLoss(m.reg_max - 1, use_dfl=self.use_dfl).to(device)
+        self.bbox_loss = BboxLoss(m.reg_max).to(device)
         self.proj = torch.arange(m.reg_max, dtype=torch.float, device=device)
 
     def preprocess(self, targets, batch_size, scale_tensor):
         """Preprocesses the target counts and matches with the input batch size to output a tensor."""
-        if targets.shape[0] == 0:
+        nl, ne = targets.shape
-            out = torch.zeros(batch_size, 0, 5, device=self.device)
+        if nl == 0:
+            out = torch.zeros(batch_size, 0, ne - 1, device=self.device)
         else:
             i = targets[:, 0]  # image index
             _, counts = i.unique(return_counts=True)
             counts = counts.to(dtype=torch.int32)
-            out = torch.zeros(batch_size, counts.max(), 5, device=self.device)
+            out = torch.zeros(batch_size, counts.max(), ne - 1, device=self.device)
             for j in range(batch_size):
                 matches = i == j
                 n = matches.sum()
@@ -592,7 +601,7 @@ class v8ClassificationLoss:
 
     def __call__(self, preds, batch):
         """Compute the classification loss between predictions and true labels."""
-        loss = torch.nn.functional.cross_entropy(preds, batch["cls"], reduction="mean")
+        loss = F.cross_entropy(preds, batch["cls"], reduction="mean")
         loss_items = loss.detach()
         return loss, loss_items
 
@@ -606,7 +615,7 @@ class v8OBBLoss(v8DetectionLoss):
         """
         super().__init__(model)
         self.assigner = RotatedTaskAlignedAssigner(topk=10, num_classes=self.nc, alpha=0.5, beta=6.0)
-        self.bbox_loss = RotatedBboxLoss(self.reg_max - 1, use_dfl=self.use_dfl).to(self.device)
+        self.bbox_loss = RotatedBboxLoss(self.reg_max).to(self.device)
 
     def preprocess(self, targets, batch_size, scale_tensor):
         """Preprocesses the target counts and matches with the input batch size to output a tensor."""