RT-ODLab/models/detectors/rtmdet_v2/matcher.py

import torch
import torch.nn as nn
import torch.nn.functional as F
from utils.box_ops import box_iou, bbox_iou


# -------------------------- Basic Functions --------------------------
def select_candidates_in_gts(xy_centers, gt_bboxes, eps=1e-9):
    """select the positive anchors's center in gt
    Args:
        xy_centers (Tensor): shape(bs*n_max_boxes, num_total_anchors, 4)
        gt_bboxes (Tensor): shape(bs, n_max_boxes, 4)
    Return:
        (Tensor): shape(bs, n_max_boxes, num_total_anchors)
    """
    n_anchors = xy_centers.size(0)
    bs, n_max_boxes, _ = gt_bboxes.size()
    _gt_bboxes = gt_bboxes.reshape([-1, 4])
    xy_centers = xy_centers.unsqueeze(0).repeat(bs * n_max_boxes, 1, 1)
    gt_bboxes_lt = _gt_bboxes[:, 0:2].unsqueeze(1).repeat(1, n_anchors, 1)
    gt_bboxes_rb = _gt_bboxes[:, 2:4].unsqueeze(1).repeat(1, n_anchors, 1)
    b_lt = xy_centers - gt_bboxes_lt
    b_rb = gt_bboxes_rb - xy_centers
    bbox_deltas = torch.cat([b_lt, b_rb], dim=-1)
    bbox_deltas = bbox_deltas.reshape([bs, n_max_boxes, n_anchors, -1])
    return (bbox_deltas.min(axis=-1)[0] > eps).to(gt_bboxes.dtype)

def select_highest_overlaps(mask_pos, overlaps, n_max_boxes):
    """if an anchor box is assigned to multiple gts,
        the one with the highest iou will be selected.
    Args:
        mask_pos (Tensor): shape(bs, n_max_boxes, num_total_anchors)
        overlaps (Tensor): shape(bs, n_max_boxes, num_total_anchors)
    Return:
        target_gt_idx (Tensor): shape(bs, num_total_anchors)
        fg_mask (Tensor): shape(bs, num_total_anchors)
        mask_pos (Tensor): shape(bs, n_max_boxes, num_total_anchors)
    """
    fg_mask = mask_pos.sum(axis=-2)
    if fg_mask.max() > 1:
        mask_multi_gts = (fg_mask.unsqueeze(1) > 1).repeat([1, n_max_boxes, 1])
        max_overlaps_idx = overlaps.argmax(axis=1)
        is_max_overlaps = F.one_hot(max_overlaps_idx, n_max_boxes)
        is_max_overlaps = is_max_overlaps.permute(0, 2, 1).to(overlaps.dtype)
        mask_pos = torch.where(mask_multi_gts, is_max_overlaps, mask_pos)
        fg_mask = mask_pos.sum(axis=-2)
    target_gt_idx = mask_pos.argmax(axis=-2)
    return target_gt_idx, fg_mask , mask_pos

def iou_calculator(box1, box2, eps=1e-9):
    """Calculate iou for batch
    Args:
        box1 (Tensor): shape(bs, n_max_boxes, 1, 4)
        box2 (Tensor): shape(bs, 1, num_total_anchors, 4)
    Return:
        (Tensor): shape(bs, n_max_boxes, num_total_anchors)
    """
    box1 = box1.unsqueeze(2)  # [N, M1, 4] -> [N, M1, 1, 4]
    box2 = box2.unsqueeze(1)  # [N, M2, 4] -> [N, 1, M2, 4]
    px1y1, px2y2 = box1[:, :, :, 0:2], box1[:, :, :, 2:4]
    gx1y1, gx2y2 = box2[:, :, :, 0:2], box2[:, :, :, 2:4]
    x1y1 = torch.maximum(px1y1, gx1y1)
    x2y2 = torch.minimum(px2y2, gx2y2)
    overlap = (x2y2 - x1y1).clip(0).prod(-1)
    area1 = (px2y2 - px1y1).clip(0).prod(-1)
    area2 = (gx2y2 - gx1y1).clip(0).prod(-1)
    union = area1 + area2 - overlap + eps

    return overlap / union


# -------------------------- Task Aligned Assigner --------------------------
class TaskAlignedAssigner(nn.Module):
    def __init__(self, topk=10, alpha=0.5, beta=6.0, eps=1e-9, num_classes=80):
        super(TaskAlignedAssigner, self).__init__()
        self.topk = topk
        self.num_classes = num_classes
        self.bg_idx = num_classes
        self.alpha = alpha
        self.beta = beta
        self.eps = eps

    @torch.no_grad()
    def forward(self,
                pd_scores,
                pd_bboxes,
                anc_points,
                gt_labels,
                gt_bboxes):
        """This code referenced to
           https://github.com/Nioolek/PPYOLOE_pytorch/blob/master/ppyoloe/assigner/tal_assigner.py
        Args:
            pd_scores (Tensor): shape(bs, num_total_anchors, num_classes)
            pd_bboxes (Tensor): shape(bs, num_total_anchors, 4)
            anc_points (Tensor): shape(num_total_anchors, 2)
            gt_labels (Tensor): shape(bs, n_max_boxes, 1)
            gt_bboxes (Tensor): shape(bs, n_max_boxes, 4)
        Returns:
            target_labels (Tensor): shape(bs, num_total_anchors)
            target_bboxes (Tensor): shape(bs, num_total_anchors, 4)
            target_scores (Tensor): shape(bs, num_total_anchors, num_classes)
            fg_mask (Tensor): shape(bs, num_total_anchors)
        """
        self.bs = pd_scores.size(0)
        self.n_max_boxes = gt_bboxes.size(1)

        mask_pos, align_metric, overlaps = self.get_pos_mask(
            pd_scores, pd_bboxes, gt_labels, gt_bboxes, anc_points)

        target_gt_idx, fg_mask, mask_pos = select_highest_overlaps(
            mask_pos, overlaps, self.n_max_boxes)

        # assigned target
        target_labels, target_bboxes, target_scores = self.get_targets(
            gt_labels, gt_bboxes, target_gt_idx, fg_mask)

        # normalize
        align_metric *= mask_pos
        pos_align_metrics = align_metric.amax(axis=-1, keepdim=True)  # b, max_num_obj
        pos_overlaps = (overlaps * mask_pos).amax(axis=-1, keepdim=True)  # b, max_num_obj
        norm_align_metric = (align_metric * pos_overlaps / (pos_align_metrics + self.eps)).amax(-2).unsqueeze(-1)
        target_scores = target_scores * norm_align_metric

        return target_labels, target_bboxes, target_scores, fg_mask.bool(), target_gt_idx


    def get_pos_mask(self, pd_scores, pd_bboxes, gt_labels, gt_bboxes, anc_points):
        # get anchor_align metric, (b, max_num_obj, h*w)
        align_metric, overlaps = self.get_box_metrics(pd_scores, pd_bboxes, gt_labels, gt_bboxes)
        # get in_gts mask, (b, max_num_obj, h*w)
        mask_in_gts = select_candidates_in_gts(anc_points, gt_bboxes)
        # get topk_metric mask, (b, max_num_obj, h*w)
        mask_topk = self.select_topk_candidates(align_metric * mask_in_gts)
        # merge all mask to a final mask, (b, max_num_obj, h*w)
        mask_pos = mask_topk * mask_in_gts

        return mask_pos, align_metric, overlaps


    def get_box_metrics(self, pd_scores, pd_bboxes, gt_labels, gt_bboxes):
        ind = torch.zeros([2, self.bs, self.n_max_boxes], dtype=torch.long)  # 2, b, max_num_obj
        ind[0] = torch.arange(end=self.bs).view(-1, 1).repeat(1, self.n_max_boxes)  # b, max_num_obj
        ind[1] = gt_labels.long().squeeze(-1)  # b, max_num_obj
        # get the scores of each grid for each gt cls
        bbox_scores = pd_scores[ind[0], :, ind[1]]  # b, max_num_obj, h*w

        overlaps = bbox_iou(gt_bboxes.unsqueeze(2), pd_bboxes.unsqueeze(1), xywh=False).squeeze(3).clamp(0)
        align_metric = bbox_scores.pow(self.alpha) * overlaps.pow(self.beta)

        return align_metric, overlaps


    def select_topk_candidates(self, metrics, largest=True):
        """
        Args:
            metrics: (b, max_num_obj, h*w).
            topk_mask: (b, max_num_obj, topk) or None
        """

        num_anchors = metrics.shape[-1]  # h*w
        # (b, max_num_obj, topk)
        topk_metrics, topk_idxs = torch.topk(metrics, self.topk, dim=-1, largest=largest)
        topk_mask = (topk_metrics.max(-1, keepdim=True)[0] > self.eps).tile([1, 1, self.topk])
        # (b, max_num_obj, topk)
        topk_idxs[~topk_mask] = 0
        # (b, max_num_obj, topk, h*w) -> (b, max_num_obj, h*w)
        is_in_topk = F.one_hot(topk_idxs, num_anchors).sum(-2)
        # filter invalid bboxes
        is_in_topk = torch.where(is_in_topk > 1, 0, is_in_topk)
        return is_in_topk.to(metrics.dtype)


    def get_targets(self, gt_labels, gt_bboxes, target_gt_idx, fg_mask):
        """
        Args:
            gt_labels: (b, max_num_obj, 1)
            gt_bboxes: (b, max_num_obj, 4)
            target_gt_idx: (b, h*w)
            fg_mask: (b, h*w)
        """

        # assigned target labels, (b, 1)
        batch_ind = torch.arange(end=self.bs, dtype=torch.int64, device=gt_labels.device)[..., None]
        target_gt_idx = target_gt_idx + batch_ind * self.n_max_boxes  # (b, h*w)
        target_labels = gt_labels.long().flatten()[target_gt_idx]  # (b, h*w)

        # assigned target boxes, (b, max_num_obj, 4) -> (b, h*w)
        target_bboxes = gt_bboxes.view(-1, 4)[target_gt_idx]

        # assigned target scores
        target_labels.clamp(0)
        target_scores = F.one_hot(target_labels, self.num_classes)  # (b, h*w, 80)
        fg_scores_mask = fg_mask[:, :, None].repeat(1, 1, self.num_classes)  # (b, h*w, 80)
        target_scores = torch.where(fg_scores_mask > 0, target_scores, 0)

        return target_labels, target_bboxes, target_scores
    

# -------------------------- Aligned SimOTA Assigner --------------------------
class AlignedSimOTA(object):
    """
        This code referenced to https://github.com/Megvii-BaseDetection/YOLOX/blob/main/yolox/models/yolo_head.py
    """
    def __init__(self, num_classes, center_sampling_radius, topk_candidate ):
        self.num_classes = num_classes
        self.center_sampling_radius = center_sampling_radius
        self.topk_candidate = topk_candidate


    @torch.no_grad()
    def __call__(self, 
                 fpn_strides, 
                 anchors, 
                 pred_cls, 
                 pred_box, 
                 tgt_labels,
                 tgt_bboxes):
        # [M,]
        strides_tensor = torch.cat([torch.ones_like(anchor_i[:, 0]) * stride_i
                                for stride_i, anchor_i in zip(fpn_strides, anchors)], dim=-1)
        # List[F, M, 2] -> [M, 2]
        anchors = torch.cat(anchors, dim=0)
        num_anchor = anchors.shape[0]        
        num_gt = len(tgt_labels)

        # ----------------------- Find inside points -----------------------
        fg_mask, is_in_boxes_and_center = self.get_in_boxes_info(
            tgt_bboxes, anchors, strides_tensor, num_anchor, num_gt)
        cls_preds = pred_cls[fg_mask].float()   # [Mp, C]
        box_preds = pred_box[fg_mask].float()   # [Mp, 4]

        # ----------------------- Reg cost -----------------------
        pair_wise_ious, _ = box_iou(tgt_bboxes, box_preds)      # [N, Mp]
        reg_cost = -torch.log(pair_wise_ious + 1e-8)            # [N, Mp]

        # ----------------------- Cls cost -----------------------
        with torch.cuda.amp.autocast(enabled=False):
            # [Mp, C] -> [N, Mp, C]
            score_preds = cls_preds.sigmoid_().unsqueeze(0).repeat(num_gt, 1, 1)
            # prepare cls_target
            cls_targets = F.one_hot(tgt_labels.long(), self.num_classes).float()
            cls_targets = cls_targets.unsqueeze(1).repeat(1, score_preds.size(1), 1)
            # [N, Mp]
            cls_cost = F.binary_cross_entropy(score_preds, cls_targets, reduction="none").sum(-1)
        del score_preds

        #----------------------- Dynamic K-Matching -----------------------
        cost_matrix = (
            cls_cost
            + 3.0 * reg_cost
            + 100000.0 * (~is_in_boxes_and_center)
        ) # [N, Mp]

        (
            assigned_labels,         # [num_fg,]
            assigned_ious,           # [num_fg,]
            assigned_indexs,         # [num_fg,]
        ) = self.dynamic_k_matching(
            cost_matrix,
            pair_wise_ious,
            tgt_labels,
            num_gt,
            fg_mask
            )
        del cls_cost, cost_matrix, pair_wise_ious, reg_cost

        return fg_mask, assigned_labels, assigned_ious, assigned_indexs


    def get_in_boxes_info(
        self,
        gt_bboxes,   # [N, 4]
        anchors,     # [M, 2]
        strides,     # [M,]
        num_anchors, # M
        num_gt,      # N
        ):
        # anchor center
        x_centers = anchors[:, 0]
        y_centers = anchors[:, 1]

        # [M,] -> [1, M] -> [N, M]
        x_centers = x_centers.unsqueeze(0).repeat(num_gt, 1)
        y_centers = y_centers.unsqueeze(0).repeat(num_gt, 1)

        # [N,] -> [N, 1] -> [N, M]
        gt_bboxes_l = gt_bboxes[:, 0].unsqueeze(1).repeat(1, num_anchors) # x1
        gt_bboxes_t = gt_bboxes[:, 1].unsqueeze(1).repeat(1, num_anchors) # y1
        gt_bboxes_r = gt_bboxes[:, 2].unsqueeze(1).repeat(1, num_anchors) # x2
        gt_bboxes_b = gt_bboxes[:, 3].unsqueeze(1).repeat(1, num_anchors) # y2

        b_l = x_centers - gt_bboxes_l
        b_r = gt_bboxes_r - x_centers
        b_t = y_centers - gt_bboxes_t
        b_b = gt_bboxes_b - y_centers
        bbox_deltas = torch.stack([b_l, b_t, b_r, b_b], 2)

        is_in_boxes = bbox_deltas.min(dim=-1).values > 0.0
        is_in_boxes_all = is_in_boxes.sum(dim=0) > 0
        # in fixed center
        center_radius = self.center_sampling_radius

        # [N, 2]
        gt_centers = (gt_bboxes[:, :2] + gt_bboxes[:, 2:]) * 0.5
        
        # [1, M]
        center_radius_ = center_radius * strides.unsqueeze(0)

        gt_bboxes_l = gt_centers[:, 0].unsqueeze(1).repeat(1, num_anchors) - center_radius_ # x1
        gt_bboxes_t = gt_centers[:, 1].unsqueeze(1).repeat(1, num_anchors) - center_radius_ # y1
        gt_bboxes_r = gt_centers[:, 0].unsqueeze(1).repeat(1, num_anchors) + center_radius_ # x2
        gt_bboxes_b = gt_centers[:, 1].unsqueeze(1).repeat(1, num_anchors) + center_radius_ # y2

        c_l = x_centers - gt_bboxes_l
        c_r = gt_bboxes_r - x_centers
        c_t = y_centers - gt_bboxes_t
        c_b = gt_bboxes_b - y_centers
        center_deltas = torch.stack([c_l, c_t, c_r, c_b], 2)
        is_in_centers = center_deltas.min(dim=-1).values > 0.0
        is_in_centers_all = is_in_centers.sum(dim=0) > 0

        # in boxes and in centers
        is_in_boxes_anchor = is_in_boxes_all | is_in_centers_all

        is_in_boxes_and_center = (
            is_in_boxes[:, is_in_boxes_anchor] & is_in_centers[:, is_in_boxes_anchor]
        )
        return is_in_boxes_anchor, is_in_boxes_and_center
    
    
    def dynamic_k_matching(
        self, 
        cost, 
        pair_wise_ious, 
        gt_classes, 
        num_gt, 
        fg_mask
        ):
        # Dynamic K
        # ---------------------------------------------------------------
        matching_matrix = torch.zeros_like(cost, dtype=torch.uint8)

        ious_in_boxes_matrix = pair_wise_ious
        n_candidate_k = min(self.topk_candidate, ious_in_boxes_matrix.size(1))
        topk_ious, _ = torch.topk(ious_in_boxes_matrix, n_candidate_k, dim=1)
        dynamic_ks = torch.clamp(topk_ious.sum(1).int(), min=1)
        dynamic_ks = dynamic_ks.tolist()
        for gt_idx in range(num_gt):
            _, pos_idx = torch.topk(
                cost[gt_idx], k=dynamic_ks[gt_idx], largest=False
            )
            matching_matrix[gt_idx][pos_idx] = 1

        del topk_ious, dynamic_ks, pos_idx

        anchor_matching_gt = matching_matrix.sum(0)
        if (anchor_matching_gt > 1).sum() > 0:
            _, cost_argmin = torch.min(cost[:, anchor_matching_gt > 1], dim=0)
            matching_matrix[:, anchor_matching_gt > 1] *= 0
            matching_matrix[cost_argmin, anchor_matching_gt > 1] = 1
        fg_mask_inboxes = matching_matrix.sum(0) > 0

        fg_mask[fg_mask.clone()] = fg_mask_inboxes

        assigned_indexs = matching_matrix[:, fg_mask_inboxes].argmax(0)
        assigned_labels = gt_classes[assigned_indexs]

        assigned_ious = (matching_matrix * pair_wise_ious).sum(0)[
            fg_mask_inboxes
        ]
        return assigned_labels, assigned_ious, assigned_indexs