Salta para o conteúdo

Referência para ultralytics/models/utils/ops.py

Nota

Este ficheiro está disponível em https://github.com/ultralytics/ ultralytics/blob/main/ ultralytics/models/utils/ops .py. Se detectares um problema, por favor ajuda a corrigi-lo contribuindo com um Pull Request 🛠️. Obrigado 🙏!



ultralytics.models.utils.ops.HungarianMatcher

Bases: Module

Um módulo que implementa o HungarianMatcher, que é um módulo diferenciável para resolver o problema de atribuição de uma forma de ponta a ponta.

O HungarianMatcher efectua uma atribuição óptima sobre as caixas delimitadoras previstas e verdadeiras utilizando uma função de custo que considera as pontuações de classificação, as coordenadas da caixa delimitadora e, opcionalmente, as previsões da máscara.

Atributos:

Nome Tipo Descrição
cost_gain dict

Dicionário de coeficientes de custo: 'class', 'bbox', 'giou', 'mask' e 'dice'.

use_fl bool

Indica se deve ser utilizada a Perda Focal para o cálculo do custo de classificação.

with_mask bool

Indica se o modelo faz previsões de máscaras.

num_sample_points int

O número de pontos de amostragem utilizados no cálculo do custo da máscara.

alpha float

O fator alfa no cálculo da perda focal.

gamma float

O fator gama no cálculo da perda focal.

Métodos:

Nome Descrição
forward

Calcula a atribui entre as previsões e as verdades fundamentais para um lote.

_cost_mask

Calcula o custo da máscara e o custo do dado se as máscaras forem previstas.

Código fonte em ultralytics/models/utils/ops.py
class HungarianMatcher(nn.Module):
    """
    A module implementing the HungarianMatcher, which is a differentiable module to solve the assignment problem in an
    end-to-end fashion.

    HungarianMatcher performs optimal assignment over the predicted and ground truth bounding boxes using a cost
    function that considers classification scores, bounding box coordinates, and optionally, mask predictions.

    Attributes:
        cost_gain (dict): Dictionary of cost coefficients: 'class', 'bbox', 'giou', 'mask', and 'dice'.
        use_fl (bool): Indicates whether to use Focal Loss for the classification cost calculation.
        with_mask (bool): Indicates whether the model makes mask predictions.
        num_sample_points (int): The number of sample points used in mask cost calculation.
        alpha (float): The alpha factor in Focal Loss calculation.
        gamma (float): The gamma factor in Focal Loss calculation.

    Methods:
        forward(pred_bboxes, pred_scores, gt_bboxes, gt_cls, gt_groups, masks=None, gt_mask=None): Computes the
            assignment between predictions and ground truths for a batch.
        _cost_mask(bs, num_gts, masks=None, gt_mask=None): Computes the mask cost and dice cost if masks are predicted.
    """

    def __init__(self, cost_gain=None, use_fl=True, with_mask=False, num_sample_points=12544, alpha=0.25, gamma=2.0):
        """Initializes HungarianMatcher with cost coefficients, Focal Loss, mask prediction, sample points, and alpha
        gamma factors.
        """
        super().__init__()
        if cost_gain is None:
            cost_gain = {"class": 1, "bbox": 5, "giou": 2, "mask": 1, "dice": 1}
        self.cost_gain = cost_gain
        self.use_fl = use_fl
        self.with_mask = with_mask
        self.num_sample_points = num_sample_points
        self.alpha = alpha
        self.gamma = gamma

    def forward(self, pred_bboxes, pred_scores, gt_bboxes, gt_cls, gt_groups, masks=None, gt_mask=None):
        """
        Forward pass for HungarianMatcher. This function computes costs based on prediction and ground truth
        (classification cost, L1 cost between boxes and GIoU cost between boxes) and finds the optimal matching between
        predictions and ground truth based on these costs.

        Args:
            pred_bboxes (Tensor): Predicted bounding boxes with shape [batch_size, num_queries, 4].
            pred_scores (Tensor): Predicted scores with shape [batch_size, num_queries, num_classes].
            gt_cls (torch.Tensor): Ground truth classes with shape [num_gts, ].
            gt_bboxes (torch.Tensor): Ground truth bounding boxes with shape [num_gts, 4].
            gt_groups (List[int]): List of length equal to batch size, containing the number of ground truths for
                each image.
            masks (Tensor, optional): Predicted masks with shape [batch_size, num_queries, height, width].
                Defaults to None.
            gt_mask (List[Tensor], optional): List of ground truth masks, each with shape [num_masks, Height, Width].
                Defaults to None.

        Returns:
            (List[Tuple[Tensor, Tensor]]): A list of size batch_size, each element is a tuple (index_i, index_j), where:
                - index_i is the tensor of indices of the selected predictions (in order)
                - index_j is the tensor of indices of the corresponding selected ground truth targets (in order)
                For each batch element, it holds:
                    len(index_i) = len(index_j) = min(num_queries, num_target_boxes)
        """

        bs, nq, nc = pred_scores.shape

        if sum(gt_groups) == 0:
            return [(torch.tensor([], dtype=torch.long), torch.tensor([], dtype=torch.long)) for _ in range(bs)]

        # We flatten to compute the cost matrices in a batch
        # [batch_size * num_queries, num_classes]
        pred_scores = pred_scores.detach().view(-1, nc)
        pred_scores = F.sigmoid(pred_scores) if self.use_fl else F.softmax(pred_scores, dim=-1)
        # [batch_size * num_queries, 4]
        pred_bboxes = pred_bboxes.detach().view(-1, 4)

        # Compute the classification cost
        pred_scores = pred_scores[:, gt_cls]
        if self.use_fl:
            neg_cost_class = (1 - self.alpha) * (pred_scores**self.gamma) * (-(1 - pred_scores + 1e-8).log())
            pos_cost_class = self.alpha * ((1 - pred_scores) ** self.gamma) * (-(pred_scores + 1e-8).log())
            cost_class = pos_cost_class - neg_cost_class
        else:
            cost_class = -pred_scores

        # Compute the L1 cost between boxes
        cost_bbox = (pred_bboxes.unsqueeze(1) - gt_bboxes.unsqueeze(0)).abs().sum(-1)  # (bs*num_queries, num_gt)

        # Compute the GIoU cost between boxes, (bs*num_queries, num_gt)
        cost_giou = 1.0 - bbox_iou(pred_bboxes.unsqueeze(1), gt_bboxes.unsqueeze(0), xywh=True, GIoU=True).squeeze(-1)

        # Final cost matrix
        C = (
            self.cost_gain["class"] * cost_class
            + self.cost_gain["bbox"] * cost_bbox
            + self.cost_gain["giou"] * cost_giou
        )
        # Compute the mask cost and dice cost
        if self.with_mask:
            C += self._cost_mask(bs, gt_groups, masks, gt_mask)

        # Set invalid values (NaNs and infinities) to 0 (fixes ValueError: matrix contains invalid numeric entries)
        C[C.isnan() | C.isinf()] = 0.0

        C = C.view(bs, nq, -1).cpu()
        indices = [linear_sum_assignment(c[i]) for i, c in enumerate(C.split(gt_groups, -1))]
        gt_groups = torch.as_tensor([0, *gt_groups[:-1]]).cumsum_(0)  # (idx for queries, idx for gt)
        return [
            (torch.tensor(i, dtype=torch.long), torch.tensor(j, dtype=torch.long) + gt_groups[k])
            for k, (i, j) in enumerate(indices)
        ]

__init__(cost_gain=None, use_fl=True, with_mask=False, num_sample_points=12544, alpha=0.25, gamma=2.0)

Inicializa o HungarianMatcher com coeficientes de custo, perda focal, previsão de máscara, pontos de amostragem e factores alfa factores gama.

Código fonte em ultralytics/models/utils/ops.py
def __init__(self, cost_gain=None, use_fl=True, with_mask=False, num_sample_points=12544, alpha=0.25, gamma=2.0):
    """Initializes HungarianMatcher with cost coefficients, Focal Loss, mask prediction, sample points, and alpha
    gamma factors.
    """
    super().__init__()
    if cost_gain is None:
        cost_gain = {"class": 1, "bbox": 5, "giou": 2, "mask": 1, "dice": 1}
    self.cost_gain = cost_gain
    self.use_fl = use_fl
    self.with_mask = with_mask
    self.num_sample_points = num_sample_points
    self.alpha = alpha
    self.gamma = gamma

forward(pred_bboxes, pred_scores, gt_bboxes, gt_cls, gt_groups, masks=None, gt_mask=None)

Passa à frente para o HungarianMatcher. Esta função calcula os custos com base na previsão e na verdade terrestre (custo de classificação, custo L1 entre caixas e custo GIoU entre caixas) e encontra a correspondência óptima entre encontra a correspondência óptima entre as previsões e a verdade terrestre com base nestes custos.

Parâmetros:

Nome Tipo Descrição Predefinição
pred_bboxes Tensor

Caixas delimitadoras previstas com a forma [batch_size, num_queries, 4].

necessário
pred_scores Tensor

Pontuações previstas com a forma [batch_size, num_queries, num_classes].

necessário
gt_cls Tensor

Classes de verdade terrestre com forma [num_gts, ].

necessário
gt_bboxes Tensor

Caixas delimitadoras da verdade terrestre com a forma [num_gts, 4].

necessário
gt_groups List[int]

Lista de comprimento igual ao tamanho do lote, contendo o número de verdades fundamentais para cada imagem.

necessário
masks Tensor

Máscaras previstas com a forma [batch_size, num_queries, height, width]. O padrão é Nenhum.

None
gt_mask List[Tensor]

Lista de máscaras de verdade terrestre, cada uma com a forma [num_masks, Height, Width]. A predefinição é Nenhum.

None

Devolve:

Tipo Descrição
List[Tuple[Tensor, Tensor]]

Uma lista de tamanho batch_size, cada elemento é uma tupla (index_i, index_j), em que: - index_i é o tensor de índices das previsões seleccionadas (por ordem) - índice_j é o tensor de índices dos alvos da verdade terrestre seleccionados correspondentes (por ordem) Para cada elemento do lote, mantém-se: len(index_i) = len(index_j) = min(num_queries, num_target_boxes)

Código fonte em ultralytics/models/utils/ops.py
def forward(self, pred_bboxes, pred_scores, gt_bboxes, gt_cls, gt_groups, masks=None, gt_mask=None):
    """
    Forward pass for HungarianMatcher. This function computes costs based on prediction and ground truth
    (classification cost, L1 cost between boxes and GIoU cost between boxes) and finds the optimal matching between
    predictions and ground truth based on these costs.

    Args:
        pred_bboxes (Tensor): Predicted bounding boxes with shape [batch_size, num_queries, 4].
        pred_scores (Tensor): Predicted scores with shape [batch_size, num_queries, num_classes].
        gt_cls (torch.Tensor): Ground truth classes with shape [num_gts, ].
        gt_bboxes (torch.Tensor): Ground truth bounding boxes with shape [num_gts, 4].
        gt_groups (List[int]): List of length equal to batch size, containing the number of ground truths for
            each image.
        masks (Tensor, optional): Predicted masks with shape [batch_size, num_queries, height, width].
            Defaults to None.
        gt_mask (List[Tensor], optional): List of ground truth masks, each with shape [num_masks, Height, Width].
            Defaults to None.

    Returns:
        (List[Tuple[Tensor, Tensor]]): A list of size batch_size, each element is a tuple (index_i, index_j), where:
            - index_i is the tensor of indices of the selected predictions (in order)
            - index_j is the tensor of indices of the corresponding selected ground truth targets (in order)
            For each batch element, it holds:
                len(index_i) = len(index_j) = min(num_queries, num_target_boxes)
    """

    bs, nq, nc = pred_scores.shape

    if sum(gt_groups) == 0:
        return [(torch.tensor([], dtype=torch.long), torch.tensor([], dtype=torch.long)) for _ in range(bs)]

    # We flatten to compute the cost matrices in a batch
    # [batch_size * num_queries, num_classes]
    pred_scores = pred_scores.detach().view(-1, nc)
    pred_scores = F.sigmoid(pred_scores) if self.use_fl else F.softmax(pred_scores, dim=-1)
    # [batch_size * num_queries, 4]
    pred_bboxes = pred_bboxes.detach().view(-1, 4)

    # Compute the classification cost
    pred_scores = pred_scores[:, gt_cls]
    if self.use_fl:
        neg_cost_class = (1 - self.alpha) * (pred_scores**self.gamma) * (-(1 - pred_scores + 1e-8).log())
        pos_cost_class = self.alpha * ((1 - pred_scores) ** self.gamma) * (-(pred_scores + 1e-8).log())
        cost_class = pos_cost_class - neg_cost_class
    else:
        cost_class = -pred_scores

    # Compute the L1 cost between boxes
    cost_bbox = (pred_bboxes.unsqueeze(1) - gt_bboxes.unsqueeze(0)).abs().sum(-1)  # (bs*num_queries, num_gt)

    # Compute the GIoU cost between boxes, (bs*num_queries, num_gt)
    cost_giou = 1.0 - bbox_iou(pred_bboxes.unsqueeze(1), gt_bboxes.unsqueeze(0), xywh=True, GIoU=True).squeeze(-1)

    # Final cost matrix
    C = (
        self.cost_gain["class"] * cost_class
        + self.cost_gain["bbox"] * cost_bbox
        + self.cost_gain["giou"] * cost_giou
    )
    # Compute the mask cost and dice cost
    if self.with_mask:
        C += self._cost_mask(bs, gt_groups, masks, gt_mask)

    # Set invalid values (NaNs and infinities) to 0 (fixes ValueError: matrix contains invalid numeric entries)
    C[C.isnan() | C.isinf()] = 0.0

    C = C.view(bs, nq, -1).cpu()
    indices = [linear_sum_assignment(c[i]) for i, c in enumerate(C.split(gt_groups, -1))]
    gt_groups = torch.as_tensor([0, *gt_groups[:-1]]).cumsum_(0)  # (idx for queries, idx for gt)
    return [
        (torch.tensor(i, dtype=torch.long), torch.tensor(j, dtype=torch.long) + gt_groups[k])
        for k, (i, j) in enumerate(indices)
    ]



ultralytics.models.utils.ops.get_cdn_group(batch, num_classes, num_queries, class_embed, num_dn=100, cls_noise_ratio=0.5, box_noise_scale=1.0, training=False)

Obtém um grupo de treino de redução de ruído contrastivo. Esta função cria um grupo de treino de redução de ruído contrastivo com amostras positivas amostras positivas e negativas das verdades fundamentais (gt). Aplica ruído às etiquetas de classe e às coordenadas da caixa delimitadora, e devolve as etiquetas modificadas, as caixas delimitadoras, a máscara de atenção e a meta-informação.

Parâmetros:

Nome Tipo Descrição Predefinição
batch dict

Um dict que inclui 'gt_cls' (torch.Tensor com forma [num_gts, ]), 'gt_bboxes' (torch.Tensor com a forma [num_gts, 4]), 'gt_groups' (List(int)) que é uma lista de tamanho de lote que indica o número de gts de cada imagem.

necessário
num_classes int

Número de aulas.

necessário
num_queries int

Número de consultas.

necessário
class_embed Tensor

Incorpora pesos para mapear etiquetas de classe para o espaço de incorporação.

necessário
num_dn int

Número de reduções de ruído. Usa a predefinição de 100.

100
cls_noise_ratio float

Rácio de ruído para etiquetas de classe. Predefine-se como 0,5.

0.5
box_noise_scale float

Escala de ruído para as coordenadas da caixa delimitadora. Usa a predefinição de 1,0.

1.0
training bool

Se estiver no modo de treino. Usa a predefinição Falso.

False

Devolve:

Tipo Descrição
Tuple[Optional[Tensor], Optional[Tensor], Optional[Tensor], Optional[Dict]]

As class embeddings modificadas, modificadas, caixas delimitadoras, máscara de atenção e meta-informação para denoising. Se não estiveres em modo de treino ou se 'num_dn' for menor ou igual a 0, a função devolve None para todos os elementos da tupla.

Código fonte em ultralytics/models/utils/ops.py
def get_cdn_group(
    batch, num_classes, num_queries, class_embed, num_dn=100, cls_noise_ratio=0.5, box_noise_scale=1.0, training=False
):
    """
    Get contrastive denoising training group. This function creates a contrastive denoising training group with positive
    and negative samples from the ground truths (gt). It applies noise to the class labels and bounding box coordinates,
    and returns the modified labels, bounding boxes, attention mask and meta information.

    Args:
        batch (dict): A dict that includes 'gt_cls' (torch.Tensor with shape [num_gts, ]), 'gt_bboxes'
            (torch.Tensor with shape [num_gts, 4]), 'gt_groups' (List(int)) which is a list of batch size length
            indicating the number of gts of each image.
        num_classes (int): Number of classes.
        num_queries (int): Number of queries.
        class_embed (torch.Tensor): Embedding weights to map class labels to embedding space.
        num_dn (int, optional): Number of denoising. Defaults to 100.
        cls_noise_ratio (float, optional): Noise ratio for class labels. Defaults to 0.5.
        box_noise_scale (float, optional): Noise scale for bounding box coordinates. Defaults to 1.0.
        training (bool, optional): If it's in training mode. Defaults to False.

    Returns:
        (Tuple[Optional[Tensor], Optional[Tensor], Optional[Tensor], Optional[Dict]]): The modified class embeddings,
            bounding boxes, attention mask and meta information for denoising. If not in training mode or 'num_dn'
            is less than or equal to 0, the function returns None for all elements in the tuple.
    """

    if (not training) or num_dn <= 0:
        return None, None, None, None
    gt_groups = batch["gt_groups"]
    total_num = sum(gt_groups)
    max_nums = max(gt_groups)
    if max_nums == 0:
        return None, None, None, None

    num_group = num_dn // max_nums
    num_group = 1 if num_group == 0 else num_group
    # Pad gt to max_num of a batch
    bs = len(gt_groups)
    gt_cls = batch["cls"]  # (bs*num, )
    gt_bbox = batch["bboxes"]  # bs*num, 4
    b_idx = batch["batch_idx"]

    # Each group has positive and negative queries.
    dn_cls = gt_cls.repeat(2 * num_group)  # (2*num_group*bs*num, )
    dn_bbox = gt_bbox.repeat(2 * num_group, 1)  # 2*num_group*bs*num, 4
    dn_b_idx = b_idx.repeat(2 * num_group).view(-1)  # (2*num_group*bs*num, )

    # Positive and negative mask
    # (bs*num*num_group, ), the second total_num*num_group part as negative samples
    neg_idx = torch.arange(total_num * num_group, dtype=torch.long, device=gt_bbox.device) + num_group * total_num

    if cls_noise_ratio > 0:
        # Half of bbox prob
        mask = torch.rand(dn_cls.shape) < (cls_noise_ratio * 0.5)
        idx = torch.nonzero(mask).squeeze(-1)
        # Randomly put a new one here
        new_label = torch.randint_like(idx, 0, num_classes, dtype=dn_cls.dtype, device=dn_cls.device)
        dn_cls[idx] = new_label

    if box_noise_scale > 0:
        known_bbox = xywh2xyxy(dn_bbox)

        diff = (dn_bbox[..., 2:] * 0.5).repeat(1, 2) * box_noise_scale  # 2*num_group*bs*num, 4

        rand_sign = torch.randint_like(dn_bbox, 0, 2) * 2.0 - 1.0
        rand_part = torch.rand_like(dn_bbox)
        rand_part[neg_idx] += 1.0
        rand_part *= rand_sign
        known_bbox += rand_part * diff
        known_bbox.clip_(min=0.0, max=1.0)
        dn_bbox = xyxy2xywh(known_bbox)
        dn_bbox = torch.logit(dn_bbox, eps=1e-6)  # inverse sigmoid

    num_dn = int(max_nums * 2 * num_group)  # total denoising queries
    # class_embed = torch.cat([class_embed, torch.zeros([1, class_embed.shape[-1]], device=class_embed.device)])
    dn_cls_embed = class_embed[dn_cls]  # bs*num * 2 * num_group, 256
    padding_cls = torch.zeros(bs, num_dn, dn_cls_embed.shape[-1], device=gt_cls.device)
    padding_bbox = torch.zeros(bs, num_dn, 4, device=gt_bbox.device)

    map_indices = torch.cat([torch.tensor(range(num), dtype=torch.long) for num in gt_groups])
    pos_idx = torch.stack([map_indices + max_nums * i for i in range(num_group)], dim=0)

    map_indices = torch.cat([map_indices + max_nums * i for i in range(2 * num_group)])
    padding_cls[(dn_b_idx, map_indices)] = dn_cls_embed
    padding_bbox[(dn_b_idx, map_indices)] = dn_bbox

    tgt_size = num_dn + num_queries
    attn_mask = torch.zeros([tgt_size, tgt_size], dtype=torch.bool)
    # Match query cannot see the reconstruct
    attn_mask[num_dn:, :num_dn] = True
    # Reconstruct cannot see each other
    for i in range(num_group):
        if i == 0:
            attn_mask[max_nums * 2 * i : max_nums * 2 * (i + 1), max_nums * 2 * (i + 1) : num_dn] = True
        if i == num_group - 1:
            attn_mask[max_nums * 2 * i : max_nums * 2 * (i + 1), : max_nums * i * 2] = True
        else:
            attn_mask[max_nums * 2 * i : max_nums * 2 * (i + 1), max_nums * 2 * (i + 1) : num_dn] = True
            attn_mask[max_nums * 2 * i : max_nums * 2 * (i + 1), : max_nums * 2 * i] = True
    dn_meta = {
        "dn_pos_idx": [p.reshape(-1) for p in pos_idx.cpu().split(list(gt_groups), dim=1)],
        "dn_num_group": num_group,
        "dn_num_split": [num_dn, num_queries],
    }

    return (
        padding_cls.to(class_embed.device),
        padding_bbox.to(class_embed.device),
        attn_mask.to(class_embed.device),
        dn_meta,
    )





Criado em 2023-11-12, Atualizado em 2023-11-25
Autores: glenn-jocher (3), Laughing-q (1)