सामग्री पर जाएं

के लिए संदर्भ ultralytics/models/utils/loss.py

नोट

यह फ़ाइल यहाँ उपलब्ध है https://github.com/ultralytics/ultralytics/बूँद/मुख्य/ultralytics/models/utils/loss.py का उपयोग करें। यदि आप कोई समस्या देखते हैं तो कृपया पुल अनुरोध का योगदान करके इसे ठीक करने में मदद करें 🛠️। 🙏 धन्यवाद !



ultralytics.models.utils.loss.DETRLoss

का रूप: Module

DETR (DEtection TRansformer) हानि वर्ग। यह वर्ग गणना करता है और के लिए विभिन्न हानि घटकों को लौटाता है DETR ऑब्जेक्ट डिटेक्शन मॉडल। यह वर्गीकरण हानि, बाउंडिंग बॉक्स हानि, GIoU हानि और वैकल्पिक रूप से सहायक की गणना करता है घाटा।

विशेषताएँ:

नाम प्रकार या क़िस्‍म
nc int

कक्षाओं की संख्या.

loss_gain dict

विभिन्न हानि घटकों के लिए गुणांक।

aux_loss bool

क्या सहायक नुकसान की गणना करना है।

use_fl bool

FocalLoss का उपयोग करें या नहीं।

use_vfl bool

VarifocalLoss का उपयोग करें या नहीं।

use_uni_match bool

सहायक शाखा के लिए लेबल असाइन करने के लिए एक निश्चित परत का उपयोग करना है या नहीं।

uni_match_ind int

उपयोग करने के लिए एक परत के निश्चित सूचकांक यदि use_uni_match सत्य है।

matcher HungarianMatcher

मिलान लागत और सूचकांकों की गणना करने पर आपत्ति।

fl FocalLoss or None

फोकल लॉस ऑब्जेक्ट अगर use_fl सच है, अन्यथा कोई नहीं।

vfl VarifocalLoss or None

Varifocal हानि वस्तु यदि use_vfl सच है, अन्यथा कोई नहीं।

device device

उपकरण जिस पर टेंसर संग्रहीत किए जाते हैं।

में स्रोत कोड ultralytics/models/utils/loss.py
class DETRLoss(nn.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:
        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 `use_uni_match` is True.
        matcher (HungarianMatcher): Object to compute matching cost and indices.
        fl (FocalLoss or None): Focal Loss object if `use_fl` is True, otherwise None.
        vfl (VarifocalLoss or None): Varifocal Loss object if `use_vfl` is True, otherwise None.
        device (torch.device): Device on which tensors are stored.
    """

    def __init__(
        self, nc=80, loss_gain=None, aux_loss=True, use_fl=True, use_vfl=False, use_uni_match=False, uni_match_ind=0
    ):
        """
        DETR loss function.

        Args:
            nc (int): The number of classes.
            loss_gain (dict): The coefficient of loss.
            aux_loss (bool): If 'aux_loss = True', loss at each decoder layer are to be used.
            use_vfl (bool): Use VarifocalLoss or not.
            use_uni_match (bool): Whether to use a fixed layer to assign labels for auxiliary branch.
            uni_match_ind (int): The fixed indices of a layer.
        """
        super().__init__()

        if loss_gain is None:
            loss_gain = {"class": 1, "bbox": 5, "giou": 2, "no_object": 0.1, "mask": 1, "dice": 1}
        self.nc = nc
        self.matcher = HungarianMatcher(cost_gain={"class": 2, "bbox": 5, "giou": 2})
        self.loss_gain = loss_gain
        self.aux_loss = aux_loss
        self.fl = FocalLoss() if use_fl else None
        self.vfl = VarifocalLoss() if use_vfl else None

        self.use_uni_match = use_uni_match
        self.uni_match_ind = uni_match_ind
        self.device = None

    def _get_loss_class(self, pred_scores, targets, gt_scores, num_gts, postfix=""):
        """Computes the classification loss based on predictions, target values, and ground truth scores."""
        # Logits: [b, query, num_classes], gt_class: list[[n, 1]]
        name_class = f"loss_class{postfix}"
        bs, nq = pred_scores.shape[:2]
        # one_hot = F.one_hot(targets, self.nc + 1)[..., :-1]  # (bs, num_queries, num_classes)
        one_hot = torch.zeros((bs, nq, self.nc + 1), dtype=torch.int64, device=targets.device)
        one_hot.scatter_(2, targets.unsqueeze(-1), 1)
        one_hot = one_hot[..., :-1]
        gt_scores = gt_scores.view(bs, nq, 1) * one_hot

        if self.fl:
            if num_gts and self.vfl:
                loss_cls = self.vfl(pred_scores, gt_scores, one_hot)
            else:
                loss_cls = self.fl(pred_scores, one_hot.float())
            loss_cls /= max(num_gts, 1) / nq
        else:
            loss_cls = nn.BCEWithLogitsLoss(reduction="none")(pred_scores, gt_scores).mean(1).sum()  # YOLO CLS loss

        return {name_class: loss_cls.squeeze() * self.loss_gain["class"]}

    def _get_loss_bbox(self, pred_bboxes, gt_bboxes, postfix=""):
        """Calculates and returns the bounding box loss and GIoU loss for the predicted and ground truth bounding
        boxes.
        """
        # Boxes: [b, query, 4], gt_bbox: list[[n, 4]]
        name_bbox = f"loss_bbox{postfix}"
        name_giou = f"loss_giou{postfix}"

        loss = {}
        if len(gt_bboxes) == 0:
            loss[name_bbox] = torch.tensor(0.0, device=self.device)
            loss[name_giou] = torch.tensor(0.0, device=self.device)
            return loss

        loss[name_bbox] = self.loss_gain["bbox"] * F.l1_loss(pred_bboxes, gt_bboxes, reduction="sum") / len(gt_bboxes)
        loss[name_giou] = 1.0 - bbox_iou(pred_bboxes, gt_bboxes, xywh=True, GIoU=True)
        loss[name_giou] = loss[name_giou].sum() / len(gt_bboxes)
        loss[name_giou] = self.loss_gain["giou"] * loss[name_giou]
        return {k: v.squeeze() for k, v in loss.items()}

    # This function is for future RT-DETR Segment models
    # def _get_loss_mask(self, masks, gt_mask, match_indices, postfix=''):
    #     # masks: [b, query, h, w], gt_mask: list[[n, H, W]]
    #     name_mask = f'loss_mask{postfix}'
    #     name_dice = f'loss_dice{postfix}'
    #
    #     loss = {}
    #     if sum(len(a) for a in gt_mask) == 0:
    #         loss[name_mask] = torch.tensor(0., device=self.device)
    #         loss[name_dice] = torch.tensor(0., device=self.device)
    #         return loss
    #
    #     num_gts = len(gt_mask)
    #     src_masks, target_masks = self._get_assigned_bboxes(masks, gt_mask, match_indices)
    #     src_masks = F.interpolate(src_masks.unsqueeze(0), size=target_masks.shape[-2:], mode='bilinear')[0]
    #     # TODO: torch does not have `sigmoid_focal_loss`, but it's not urgent since we don't use mask branch for now.
    #     loss[name_mask] = self.loss_gain['mask'] * F.sigmoid_focal_loss(src_masks, target_masks,
    #                                                                     torch.tensor([num_gts], dtype=torch.float32))
    #     loss[name_dice] = self.loss_gain['dice'] * self._dice_loss(src_masks, target_masks, num_gts)
    #     return loss

    # This function is for future RT-DETR Segment models
    # @staticmethod
    # def _dice_loss(inputs, targets, num_gts):
    #     inputs = F.sigmoid(inputs).flatten(1)
    #     targets = targets.flatten(1)
    #     numerator = 2 * (inputs * targets).sum(1)
    #     denominator = inputs.sum(-1) + targets.sum(-1)
    #     loss = 1 - (numerator + 1) / (denominator + 1)
    #     return loss.sum() / num_gts

    def _get_loss_aux(
        self,
        pred_bboxes,
        pred_scores,
        gt_bboxes,
        gt_cls,
        gt_groups,
        match_indices=None,
        postfix="",
        masks=None,
        gt_mask=None,
    ):
        """Get auxiliary losses."""
        # NOTE: loss class, bbox, giou, mask, dice
        loss = torch.zeros(5 if masks is not None else 3, device=pred_bboxes.device)
        if match_indices is None and self.use_uni_match:
            match_indices = self.matcher(
                pred_bboxes[self.uni_match_ind],
                pred_scores[self.uni_match_ind],
                gt_bboxes,
                gt_cls,
                gt_groups,
                masks=masks[self.uni_match_ind] if masks is not None else None,
                gt_mask=gt_mask,
            )
        for i, (aux_bboxes, aux_scores) in enumerate(zip(pred_bboxes, pred_scores)):
            aux_masks = masks[i] if masks is not None else None
            loss_ = self._get_loss(
                aux_bboxes,
                aux_scores,
                gt_bboxes,
                gt_cls,
                gt_groups,
                masks=aux_masks,
                gt_mask=gt_mask,
                postfix=postfix,
                match_indices=match_indices,
            )
            loss[0] += loss_[f"loss_class{postfix}"]
            loss[1] += loss_[f"loss_bbox{postfix}"]
            loss[2] += loss_[f"loss_giou{postfix}"]
            # if masks is not None and gt_mask is not None:
            #     loss_ = self._get_loss_mask(aux_masks, gt_mask, match_indices, postfix)
            #     loss[3] += loss_[f'loss_mask{postfix}']
            #     loss[4] += loss_[f'loss_dice{postfix}']

        loss = {
            f"loss_class_aux{postfix}": loss[0],
            f"loss_bbox_aux{postfix}": loss[1],
            f"loss_giou_aux{postfix}": loss[2],
        }
        # if masks is not None and gt_mask is not None:
        #     loss[f'loss_mask_aux{postfix}'] = loss[3]
        #     loss[f'loss_dice_aux{postfix}'] = loss[4]
        return loss

    @staticmethod
    def _get_index(match_indices):
        """Returns batch indices, source indices, and destination indices from provided match indices."""
        batch_idx = torch.cat([torch.full_like(src, i) for i, (src, _) in enumerate(match_indices)])
        src_idx = torch.cat([src for (src, _) in match_indices])
        dst_idx = torch.cat([dst for (_, dst) in match_indices])
        return (batch_idx, src_idx), dst_idx

    def _get_assigned_bboxes(self, pred_bboxes, gt_bboxes, match_indices):
        """Assigns predicted bounding boxes to ground truth bounding boxes based on the match indices."""
        pred_assigned = torch.cat(
            [
                t[i] if len(i) > 0 else torch.zeros(0, t.shape[-1], device=self.device)
                for t, (i, _) in zip(pred_bboxes, match_indices)
            ]
        )
        gt_assigned = torch.cat(
            [
                t[j] if len(j) > 0 else torch.zeros(0, t.shape[-1], device=self.device)
                for t, (_, j) in zip(gt_bboxes, match_indices)
            ]
        )
        return pred_assigned, gt_assigned

    def _get_loss(
        self,
        pred_bboxes,
        pred_scores,
        gt_bboxes,
        gt_cls,
        gt_groups,
        masks=None,
        gt_mask=None,
        postfix="",
        match_indices=None,
    ):
        """Get losses."""
        if match_indices is None:
            match_indices = self.matcher(
                pred_bboxes, pred_scores, gt_bboxes, gt_cls, gt_groups, masks=masks, gt_mask=gt_mask
            )

        idx, gt_idx = self._get_index(match_indices)
        pred_bboxes, gt_bboxes = pred_bboxes[idx], gt_bboxes[gt_idx]

        bs, nq = pred_scores.shape[:2]
        targets = torch.full((bs, nq), self.nc, device=pred_scores.device, dtype=gt_cls.dtype)
        targets[idx] = gt_cls[gt_idx]

        gt_scores = torch.zeros([bs, nq], device=pred_scores.device)
        if len(gt_bboxes):
            gt_scores[idx] = bbox_iou(pred_bboxes.detach(), gt_bboxes, xywh=True).squeeze(-1)

        loss = {}
        loss.update(self._get_loss_class(pred_scores, targets, gt_scores, len(gt_bboxes), postfix))
        loss.update(self._get_loss_bbox(pred_bboxes, gt_bboxes, postfix))
        # if masks is not None and gt_mask is not None:
        #     loss.update(self._get_loss_mask(masks, gt_mask, match_indices, postfix))
        return loss

    def forward(self, pred_bboxes, pred_scores, batch, postfix="", **kwargs):
        """
        Args:
            pred_bboxes (torch.Tensor): [l, b, query, 4]
            pred_scores (torch.Tensor): [l, b, query, num_classes]
            batch (dict): A dict includes:
                gt_cls (torch.Tensor) with shape [num_gts, ],
                gt_bboxes (torch.Tensor): [num_gts, 4],
                gt_groups (List(int)): a list of batch size length includes the number of gts of each image.
            postfix (str): postfix of loss name.
        """
        self.device = pred_bboxes.device
        match_indices = kwargs.get("match_indices", None)
        gt_cls, gt_bboxes, gt_groups = batch["cls"], batch["bboxes"], batch["gt_groups"]

        total_loss = self._get_loss(
            pred_bboxes[-1], pred_scores[-1], gt_bboxes, gt_cls, gt_groups, postfix=postfix, match_indices=match_indices
        )

        if self.aux_loss:
            total_loss.update(
                self._get_loss_aux(
                    pred_bboxes[:-1], pred_scores[:-1], gt_bboxes, gt_cls, gt_groups, match_indices, postfix
                )
            )

        return total_loss

__init__(nc=80, loss_gain=None, aux_loss=True, use_fl=True, use_vfl=False, use_uni_match=False, uni_match_ind=0)

DETR नुकसान समारोह।

पैरामीटर:

नाम प्रकार या क़िस्‍म चूक
nc int

कक्षाओं की संख्या.

80
loss_gain dict

हानि का गुणांक।

None
aux_loss bool

यदि 'aux_loss = सत्य' है, तो प्रत्येक डिकोडर परत पर नुकसान का उपयोग किया जाना है।

True
use_vfl bool

VarifocalLoss का उपयोग करें या नहीं।

False
use_uni_match bool

सहायक शाखा के लिए लेबल असाइन करने के लिए एक निश्चित परत का उपयोग करना है या नहीं।

False
uni_match_ind int

एक परत के निश्चित सूचकांक।

0
में स्रोत कोड ultralytics/models/utils/loss.py
def __init__(
    self, nc=80, loss_gain=None, aux_loss=True, use_fl=True, use_vfl=False, use_uni_match=False, uni_match_ind=0
):
    """
    DETR loss function.

    Args:
        nc (int): The number of classes.
        loss_gain (dict): The coefficient of loss.
        aux_loss (bool): If 'aux_loss = True', loss at each decoder layer are to be used.
        use_vfl (bool): Use VarifocalLoss or not.
        use_uni_match (bool): Whether to use a fixed layer to assign labels for auxiliary branch.
        uni_match_ind (int): The fixed indices of a layer.
    """
    super().__init__()

    if loss_gain is None:
        loss_gain = {"class": 1, "bbox": 5, "giou": 2, "no_object": 0.1, "mask": 1, "dice": 1}
    self.nc = nc
    self.matcher = HungarianMatcher(cost_gain={"class": 2, "bbox": 5, "giou": 2})
    self.loss_gain = loss_gain
    self.aux_loss = aux_loss
    self.fl = FocalLoss() if use_fl else None
    self.vfl = VarifocalLoss() if use_vfl else None

    self.use_uni_match = use_uni_match
    self.uni_match_ind = uni_match_ind
    self.device = None

forward(pred_bboxes, pred_scores, batch, postfix='', **kwargs)

पैरामीटर:

नाम प्रकार या क़िस्‍म चूक
pred_bboxes Tensor

[एल, बी, क्वेरी, 4]

आवश्यक
pred_scores Tensor

[एल, बी, क्वेरी, num_classes]

आवश्यक
batch dict

एक डिक्ट में शामिल हैं: gt_cls (torch.Tensor) आकार के साथ [num_gts,], gt_bboxes (torch.Tensor): [num_gts, 4], gt_groups (सूची (int)): बैच आकार की लंबाई की सूची में प्रत्येक छवि के gts की संख्या शामिल होती है।

आवश्यक
postfix str

नुकसान के नाम का पोस्टफिक्स।

''
में स्रोत कोड ultralytics/models/utils/loss.py
def forward(self, pred_bboxes, pred_scores, batch, postfix="", **kwargs):
    """
    Args:
        pred_bboxes (torch.Tensor): [l, b, query, 4]
        pred_scores (torch.Tensor): [l, b, query, num_classes]
        batch (dict): A dict includes:
            gt_cls (torch.Tensor) with shape [num_gts, ],
            gt_bboxes (torch.Tensor): [num_gts, 4],
            gt_groups (List(int)): a list of batch size length includes the number of gts of each image.
        postfix (str): postfix of loss name.
    """
    self.device = pred_bboxes.device
    match_indices = kwargs.get("match_indices", None)
    gt_cls, gt_bboxes, gt_groups = batch["cls"], batch["bboxes"], batch["gt_groups"]

    total_loss = self._get_loss(
        pred_bboxes[-1], pred_scores[-1], gt_bboxes, gt_cls, gt_groups, postfix=postfix, match_indices=match_indices
    )

    if self.aux_loss:
        total_loss.update(
            self._get_loss_aux(
                pred_bboxes[:-1], pred_scores[:-1], gt_bboxes, gt_cls, gt_groups, match_indices, postfix
            )
        )

    return total_loss



ultralytics.models.utils.loss.RTDETRDetectionLoss

का रूप: DETRLoss

रीयल-टाइम डीपट्रैकर (RT-DETR) डिटेक्शन लॉस क्लास जो DETRLoss का विस्तार करता है।

यह वर्ग के लिए पता लगाने के नुकसान की गणना करता है RT-DETR मॉडल, जिसमें मानक पहचान हानि के साथ-साथ मेटाडेटा को डीनोइज़ करने के साथ प्रदान किए जाने पर एक अतिरिक्त डीनोइज़िंग प्रशिक्षण हानि।

में स्रोत कोड ultralytics/models/utils/loss.py
class RTDETRDetectionLoss(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.
    """

    def forward(self, preds, batch, dn_bboxes=None, dn_scores=None, dn_meta=None):
        """
        Forward pass to compute the detection loss.

        Args:
            preds (tuple): Predicted bounding boxes and scores.
            batch (dict): Batch data containing ground truth information.
            dn_bboxes (torch.Tensor, optional): Denoising bounding boxes. Default is None.
            dn_scores (torch.Tensor, optional): Denoising scores. Default is None.
            dn_meta (dict, optional): Metadata for denoising. Default is None.

        Returns:
            (dict): Dictionary containing the total loss and, if applicable, the denoising loss.
        """
        pred_bboxes, pred_scores = preds
        total_loss = super().forward(pred_bboxes, pred_scores, batch)

        # Check for denoising metadata to compute denoising training loss
        if dn_meta is not None:
            dn_pos_idx, dn_num_group = dn_meta["dn_pos_idx"], dn_meta["dn_num_group"]
            assert len(batch["gt_groups"]) == len(dn_pos_idx)

            # Get the match indices for denoising
            match_indices = self.get_dn_match_indices(dn_pos_idx, dn_num_group, batch["gt_groups"])

            # Compute the denoising training loss
            dn_loss = super().forward(dn_bboxes, dn_scores, batch, postfix="_dn", match_indices=match_indices)
            total_loss.update(dn_loss)
        else:
            # If no denoising metadata is provided, set denoising loss to zero
            total_loss.update({f"{k}_dn": torch.tensor(0.0, device=self.device) for k in total_loss.keys()})

        return total_loss

    @staticmethod
    def get_dn_match_indices(dn_pos_idx, dn_num_group, gt_groups):
        """
        Get the match indices for denoising.

        Args:
            dn_pos_idx (List[torch.Tensor]): List of tensors containing positive indices for denoising.
            dn_num_group (int): Number of denoising groups.
            gt_groups (List[int]): List of integers representing the number of ground truths for each image.

        Returns:
            (List[tuple]): List of tuples containing matched indices for denoising.
        """
        dn_match_indices = []
        idx_groups = torch.as_tensor([0, *gt_groups[:-1]]).cumsum_(0)
        for i, num_gt in enumerate(gt_groups):
            if num_gt > 0:
                gt_idx = torch.arange(end=num_gt, dtype=torch.long) + idx_groups[i]
                gt_idx = gt_idx.repeat(dn_num_group)
                assert len(dn_pos_idx[i]) == len(gt_idx), "Expected the same length, "
                f"but got {len(dn_pos_idx[i])} and {len(gt_idx)} respectively."
                dn_match_indices.append((dn_pos_idx[i], gt_idx))
            else:
                dn_match_indices.append((torch.zeros([0], dtype=torch.long), torch.zeros([0], dtype=torch.long)))
        return dn_match_indices

forward(preds, batch, dn_bboxes=None, dn_scores=None, dn_meta=None)

पता लगाने के नुकसान की गणना करने के लिए फॉरवर्ड पास।

पैरामीटर:

नाम प्रकार या क़िस्‍म चूक
preds tuple

बाउंडिंग बॉक्स और स्कोर की भविष्यवाणी की।

आवश्यक
batch dict

जमीनी सच्चाई की जानकारी वाला बैच डेटा।

आवश्यक
dn_bboxes Tensor

बाउंडिंग बॉक्स को डीनोइज़ करना। डिफ़ॉल्ट कोई नहीं है।

None
dn_scores Tensor

स्कोर को कम करना। डिफ़ॉल्ट कोई नहीं है।

None
dn_meta dict

denoising के लिए मेटाडेटा। डिफ़ॉल्ट कोई नहीं है।

None

देता:

प्रकार या क़िस्‍म
dict

शब्दकोश जिसमें कुल नुकसान और, यदि लागू हो, तो डीनोइज़िंग लॉस शामिल है।

में स्रोत कोड ultralytics/models/utils/loss.py
def forward(self, preds, batch, dn_bboxes=None, dn_scores=None, dn_meta=None):
    """
    Forward pass to compute the detection loss.

    Args:
        preds (tuple): Predicted bounding boxes and scores.
        batch (dict): Batch data containing ground truth information.
        dn_bboxes (torch.Tensor, optional): Denoising bounding boxes. Default is None.
        dn_scores (torch.Tensor, optional): Denoising scores. Default is None.
        dn_meta (dict, optional): Metadata for denoising. Default is None.

    Returns:
        (dict): Dictionary containing the total loss and, if applicable, the denoising loss.
    """
    pred_bboxes, pred_scores = preds
    total_loss = super().forward(pred_bboxes, pred_scores, batch)

    # Check for denoising metadata to compute denoising training loss
    if dn_meta is not None:
        dn_pos_idx, dn_num_group = dn_meta["dn_pos_idx"], dn_meta["dn_num_group"]
        assert len(batch["gt_groups"]) == len(dn_pos_idx)

        # Get the match indices for denoising
        match_indices = self.get_dn_match_indices(dn_pos_idx, dn_num_group, batch["gt_groups"])

        # Compute the denoising training loss
        dn_loss = super().forward(dn_bboxes, dn_scores, batch, postfix="_dn", match_indices=match_indices)
        total_loss.update(dn_loss)
    else:
        # If no denoising metadata is provided, set denoising loss to zero
        total_loss.update({f"{k}_dn": torch.tensor(0.0, device=self.device) for k in total_loss.keys()})

    return total_loss

get_dn_match_indices(dn_pos_idx, dn_num_group, gt_groups) staticmethod

denoising के लिए मैच सूचकांक प्राप्त करें।

पैरामीटर:

नाम प्रकार या क़िस्‍म चूक
dn_pos_idx List[Tensor]

डीनोइज़िंग के लिए सकारात्मक सूचकांकों वाले टेंसर की सूची।

आवश्यक
dn_num_group int

संप्रदाय समूहों की संख्या।

आवश्यक
gt_groups List[int]

प्रत्येक छवि के लिए जमीनी सत्य की संख्या का प्रतिनिधित्व करने वाले पूर्णांकों की सूची।

आवश्यक

देता:

प्रकार या क़िस्‍म
List[tuple]

डीनोइज़िंग के लिए मिलान किए गए सूचकांकों वाले टुपल्स की सूची।

में स्रोत कोड ultralytics/models/utils/loss.py
@staticmethod
def get_dn_match_indices(dn_pos_idx, dn_num_group, gt_groups):
    """
    Get the match indices for denoising.

    Args:
        dn_pos_idx (List[torch.Tensor]): List of tensors containing positive indices for denoising.
        dn_num_group (int): Number of denoising groups.
        gt_groups (List[int]): List of integers representing the number of ground truths for each image.

    Returns:
        (List[tuple]): List of tuples containing matched indices for denoising.
    """
    dn_match_indices = []
    idx_groups = torch.as_tensor([0, *gt_groups[:-1]]).cumsum_(0)
    for i, num_gt in enumerate(gt_groups):
        if num_gt > 0:
            gt_idx = torch.arange(end=num_gt, dtype=torch.long) + idx_groups[i]
            gt_idx = gt_idx.repeat(dn_num_group)
            assert len(dn_pos_idx[i]) == len(gt_idx), "Expected the same length, "
            f"but got {len(dn_pos_idx[i])} and {len(gt_idx)} respectively."
            dn_match_indices.append((dn_pos_idx[i], gt_idx))
        else:
            dn_match_indices.append((torch.zeros([0], dtype=torch.long), torch.zeros([0], dtype=torch.long)))
    return dn_match_indices





Created 2023-11-12, Updated 2024-06-02
Authors: glenn-jocher (5), Burhan-Q (1), Laughing-q (1)