Fi&`ddlmZddlZddlmZddlmZddlmZm Z ddl m Z m Z m Z ddlmZGdd ej ZGd d eZdd Zdd ZdddZddZ d ddZy)) annotationsN)LOGGER)bbox_iouprobiou) xywh2xyxyxywhr2xyxyxyxy xyxy2xywh) TORCH_1_11ceZdZdZddddgdddf dfd Zej d Zd Zd Z d Z dZ ddZ dZ ddZdZxZS)TaskAlignedAssigneraGA task-aligned assigner for object detection. This class assigns ground-truth (gt) objects to anchors based on the task-aligned metric, which combines both classification and localization information. Attributes: topk (int): The number of top candidates to consider. topk2 (int): Secondary topk value for additional filtering. num_classes (int): The number of object classes. alpha (float): The alpha parameter for the classification component of the task-aligned metric. beta (float): The beta parameter for the localization component of the task-aligned metric. stride (list): List of stride values for different feature levels. stride_val (int): The stride value used for select_candidates_in_gts. eps (float): A small value to prevent division by zero. P?g@) & .>Nct|||_|xs||_||_||_||_||_t|jdkDr|jdn|jd|_ ||_ y)aInitialize a TaskAlignedAssigner object with customizable hyperparameters. Args: topk (int, optional): The number of top candidates to consider. num_classes (int, optional): The number of object classes. alpha (float, optional): The alpha parameter for the classification component of the task-aligned metric. beta (float, optional): The beta parameter for the localization component of the task-aligned metric. stride (list, optional): List of stride values for different feature levels. eps (float, optional): A small value to prevent division by zero. topk2 (int, optional): Secondary topk value for additional filtering. rrN) super__init__topktopk2 num_classesalphabetastridelen stride_valeps) selfrrrrrr r __class__s W/var/www/pod-logistic/pod-ai/venv/lib/python3.12/site-packages/ultralytics/utils/tal.pyrzTaskAlignedAssigner.__init__sp*  ]d &   ,/ ,z.TaskAlignedAssigner.forward..is:aQTT&\:sN)shapebs n_max_boxesr+torch full_liker zeros_like_forward RuntimeErrorstrlowerrwarningcputuple) r! pd_scores pd_bboxes anc_points gt_labels gt_bboxesmask_gter* cpu_tensorsresultr+s @r#forwardzTaskAlignedAssigner.forward>sB,//!$$??1-!!   q  & 143C3CD  +  +  6!23  6!23   ==Iz9iY`a a #a&,,.0XY1:IzS\^gip0qr1quuwrr r& 4:6:::  s0C E&:E!D10*E!E& E!!E&c|j||||||\}}} |j|| |j|\} } }|j||| | \} } }||z}|j dd}| |zj dd}||z||j zz j dj d}||z}| | || j| fS)aCompute the task-aligned assignment. Args: pd_scores (torch.Tensor): Predicted classification scores with shape (bs, num_total_anchors, num_classes). pd_bboxes (torch.Tensor): Predicted bounding boxes with shape (bs, num_total_anchors, 4). anc_points (torch.Tensor): Anchor points with shape (num_total_anchors, 2). gt_labels (torch.Tensor): Ground truth labels with shape (bs, n_max_boxes, 1). gt_bboxes (torch.Tensor): Ground truth boxes with shape (bs, n_max_boxes, 4). mask_gt (torch.Tensor): Mask for valid ground truth boxes with shape (bs, n_max_boxes, 1). Returns: target_labels (torch.Tensor): Target labels with shape (bs, num_total_anchors). target_bboxes (torch.Tensor): Target bounding boxes with shape (bs, num_total_anchors, 4). target_scores (torch.Tensor): Target scores with shape (bs, num_total_anchors, num_classes). fg_mask (torch.Tensor): Foreground mask with shape (bs, num_total_anchors). target_gt_idx (torch.Tensor): Target ground truth indices with shape (bs, num_total_anchors). T)dimkeepdim) get_pos_maskselect_highest_overlapsr/ get_targetsamaxr unsqueezebool)r!r:r;r<r=r>r?mask_pos align_metricoverlaps target_gt_idxfg_mask target_labels target_bboxes target_scorespos_align_metrics pos_overlapsnorm_align_metrics r#r3zTaskAlignedAssigner._forwardls$,0+<+< y)Y G, (,,0+G+G h 0 0,, ( w 7;6F6FyR[]jls6t3 }m  (--"d-C 8+11b$1G )L8r<r? mask_in_gtsrPrQ mask_topkrOs r#rIz TaskAlignedAssigner.get_pos_masks 33J 7S !%!5!5iIW`bmpwbw!x h// WY[]_c_h_hHiHnHnHp/q {*W4x//r$c|jd}|j}tj|j|j |g|j |j}tj|j|j |g|j |j}tjd|j|j gtj} tj|jjddjd|j | d<|jd| d<|| dd d | df|||<|jdjd|j dd|} |jdjdd|d|} |j| | ||<|j|j |j|j"z} | |fS) a/Compute alignment metric given predicted and ground truth bounding boxes. Args: pd_scores (torch.Tensor): Predicted classification scores with shape (bs, num_total_anchors, num_classes). pd_bboxes (torch.Tensor): Predicted bounding boxes with shape (bs, num_total_anchors, 4). gt_labels (torch.Tensor): Ground truth labels with shape (bs, n_max_boxes, 1). gt_bboxes (torch.Tensor): Ground truth boxes with shape (bs, n_max_boxes, 4). mask_gt (torch.Tensor): Mask for valid ground truth boxes with shape (bs, n_max_boxes, h*w). Returns: align_metric (torch.Tensor): Alignment metric combining classification and localization. overlaps (torch.Tensor): IoU overlaps between predicted and ground truth boxes. rHdtyper+)rd)endrErrN)r-rNr0zerosr.r/rdr+longarangeviewr_squeezerMiou_calculationpowrr) r!r:r;r=r>r?narQ bbox_scoresindpd_boxesgt_boxesrPs r#r]z#TaskAlignedAssigner.get_box_metricss__R ,,.;;)9)92>ioo^g^n^nokk477D,<,r;s r#rlz#TaskAlignedAssigner.iou_calculations, 95tDLLRPWWXYZZr$c tj||jdd\}}|2|jddd|jkDj |}|j |dtj |jtj|j}tj|ddddddftj|j}t|jD]$}|jd|dddd||dzf|&|j |dkDd|j|jS) aSelect the top-k candidates based on the given metrics. Args: metrics (torch.Tensor): A tensor of shape (b, max_num_obj, h*w), where b is the batch size, max_num_obj is the maximum number of objects, and h*w represents the total number of anchor points. topk_mask (torch.Tensor, optional): An optional boolean tensor of shape (b, max_num_obj, topk), where topk is the number of top candidates to consider. If not provided, the top-k values are automatically computed based on the given metrics. Returns: (torch.Tensor): A tensor of shape (b, max_num_obj, h*w) containing the selected top-k candidates. rETrFlargestN)rGrrcr)r0rmaxr expand_as masked_fill_rgr-int8r+ ones_likerange scatter_add_r(rd)r!metricsr[ topk_metrics topk_idxs count_tensoronesks r#r^z*TaskAlignedAssigner.select_topk_candidatess#(**WdiiRQU"V i  %))"d);A>ITTU^_I z1-{{7== 9K[K[\ yArr2%**YM]M]^tyy! LA  % %b)Aq!a!e)O*Dd K L !!,"2A6w}}--r$ctj|jtj|jd}|||j zz}|j j|}|jd|jd|}|jdtj|jd|jd|jftj|j}|jd|jdd|dddddfjdd|j} tj | dkD|d}|||fS) a@Compute target labels, target bounding boxes, and target scores for the positive anchor points. Args: gt_labels (torch.Tensor): Ground truth labels of shape (b, max_num_obj, 1), where b is the batch size and max_num_obj is the maximum number of objects. gt_bboxes (torch.Tensor): Ground truth bounding boxes of shape (b, max_num_obj, 4). target_gt_idx (torch.Tensor): Indices of the assigned ground truth objects for positive anchor points, with shape (b, h*w), where h*w is the total number of anchor points. fg_mask (torch.Tensor): A boolean tensor of shape (b, h*w) indicating the positive (foreground) anchor points. Returns: target_labels (torch.Tensor): Target labels for positive anchor points with shape (b, h*w). target_bboxes (torch.Tensor): Target bounding boxes for positive anchor points with shape (b, h*w, 4). target_scores (torch.Tensor): Target scores for positive anchor points with shape (b, h*w, num_classes). )rfrdr+).NrErrrcreN)r0rir.int64r+r/rhflattenrjr-rvrgrscatter_rMrepeatwhere) r!r=r>rRrS batch_indrTrUrVfg_scores_masks r#rKzTaskAlignedAssigner.get_targetss7$LLTWWEKK HXHXYZcd % D4D4D(DD !(002=A "r9??2+>? N  Q   #]%8%8%;T=M=M N++ '' q-"9"9""=qA At,33Aq$:J:JK NQ$6 qI m]::r$clt|}|dddf|jdk}tj||zj tj |j |j|j|dddf|dddf<t|}|jd}|j\}} } |jdddjdd\} } tj|d| z | |dz fd j|| |d} | jd j|S) aSelect positive anchor centers within ground truth bounding boxes. Args: xy_centers (torch.Tensor): Anchor center coordinates, shape (h*w, 2). gt_bboxes (torch.Tensor): Ground truth bounding boxes, shape (b, n_boxes, 4). mask_gt (torch.Tensor): Mask for valid ground truth boxes, shape (b, n_boxes, 1). eps (float, optional): Small value for numerical stability. Returns: (torch.Tensor): Boolean mask of positive anchors, shape (b, n_boxes, h*w). Notes: - b: batch size, n_boxes: number of ground truth boxes, h: height, w: width. - Bounding box format: [x_min, y_min, x_max, y_max]. .reNrrcrErrF)r rr0rrNtensorrrdr+rr-rjchunkcatamingt_)r! xy_centersr>r?r gt_bboxes_xywhwh_mask n_anchorsr.n_boxes_ltrb bbox_deltass r#r\z,TaskAlignedAssigner.select_candidates_in_gts!s, #9- ab)DKKN:"'++ w  $ $ & LL0D0D^MbMb c 37 ## sABw n- $$Q' "GQAq)//15BiiD!1B!6Z=M8M NTUV[[\^`girtvw "&&s++r$cR|jd}|jdkDr|jddkDjd|d}|j d}t j |j|j|j}|jd|jddt j|||j}|jd}|j|jk7r||z}t j||jddj}t j |j|j|j} | jd|d|| z}|jd}|j d} | ||fS)aSelect anchor boxes with highest IoU when assigned to multiple ground truths. Args: mask_pos (torch.Tensor): Positive mask, shape (b, n_max_boxes, h*w). overlaps (torch.Tensor): IoU overlaps, shape (b, n_max_boxes, h*w). n_max_boxes (int): Maximum number of ground truth boxes. align_metric (torch.Tensor): Alignment metric for selecting best matches. Returns: target_gt_idx (torch.Tensor): Indices of assigned ground truths, shape (b, h*w). fg_mask (torch.Tensor): Foreground mask, shape (b, h*w). mask_pos (torch.Tensor): Updated positive mask, shape (b, n_max_boxes, h*w). rHrrErcTryr)sumr{rMr_argmaxr0rgr-rdr+rrfloatrrindices) r!rOrQr/rPrSmask_multi_gtsmax_overlaps_idxis_max_overlapstopk_idxrRs r#rJz+TaskAlignedAssigner.select_highest_overlaps@s],,r" ;;=1 %//2Q6>>r;PRSN'q1 #kk(..W_WfWfgO  $ $Q(8(B(B1(Eq I{{>?HMSSUHll2&G :: "'(2L$zz, TXYaa {{8>>PXP_P_`H   b"2C 8  Hll2&G + gx//r$) rintrrrrrrrlistr rr')r)__name__ __module__ __qualname____doc__rr0no_gradrCr3rIr]rlr^rKr\rJ __classcell__)r"s@r#r r s$"     >U]]_++Z$ZL04&B [.>';R,>#0r$r ceZdZdZdZdZy)RotatedTaskAlignedAssignerzSAssigns ground-truth objects to rotated bounding boxes using a task-aligned metric.cVt||jdjdS)z)Calculate IoU for rotated bounding boxes.rEr)rrkrvrws r#rlz*RotatedTaskAlignedAssigner.iou_calculationis%y),44R8??BBr$cb|j}|dddf|jdk}tj||zj tj |j |j|j|dddf|dddf<t|}|jdd\}}} } ||z } | |z } ||z } | | zjd }| | zjd }| | zjd }| | zjd }|dk\||kz|dk\z||kzS) aSelect the positive anchor center in gt for rotated bounding boxes. Args: xy_centers (torch.Tensor): Anchor center coordinates with shape (h*w, 2). gt_bboxes (torch.Tensor): Ground truth bounding boxes with shape (b, n_boxes, 5). mask_gt (torch.Tensor): Mask for valid ground truth boxes with shape (b, n_boxes, 1). Returns: (torch.Tensor): Boolean mask of positive anchors with shape (b, n_boxes, h*w). .rerrrcrrHrrE) clonerr0rrNrrrdr+r splitr)r!rr>r?gt_bboxes_clonercornersabrdabadapnorm_abnorm_ad ap_dot_ab ap_dot_ads r#r\z3RotatedTaskAlignedAssigner.select_candidates_in_gtsmsF$//+!#qs(+dkk!n<$)KK w  $ $ & LL0E0EoNdNd e C1H %% QqS!!1]]1"]- 1a U U!^7--B-'7--B-'"WMMbM) "WMMbM) Q9#78INKy\cOcddr$N)rrrrrlr\r$r#rrfs]C er$rc gg}}|J|dj|dj}}tt|D] }||}t |t r||j ddn!t||dt||df\} } tj| |||z} tj| |||z} trtj| | dntj| | \} } |jtj| | fdjdd|jtj| | zdf||| #tj |tj |fS) zGenerate anchors from features.Nrrer)rfr+rdij)indexingrErc)rdr+rr isinstancerr-rr0rir meshgridappendstackrjfullr) featsstridesgrid_cell_offset anchor_points stride_tensorrdr+irhwsxsys r# make_anchorsrsb#%r=M   !HNNE!HOO6E 3u: Y%/t% )B 88A38 DAq1urzH1urzH1urzH1urzH 99h(H=3 GD{{1gn- Kr$)g?)TrEr')r torch.Tensorrrr int | Nonereturnr)rE)rEN) rUrrrrrrFrrr) __future__rr0torch.nnnnrrrropsrr r torch_utilsr Moduler rrrrrrrr$r#rs# &55#U0"))U0p 'e!4'eT > (-2 $$$$ $  $r$